AI Strategy - Tygart Media

Category: AI Strategy

AI strategy for operators: deploy Claude, automate real workflows, and build AI-native systems that compound. Field notes and playbooks from Tygart Media.

  • Is Claude Smarter Than ChatGPT? 2026 Benchmark Guide

    Is Claude Smarter Than ChatGPT? 2026 Benchmark Guide

    Last refreshed: May 15, 2026

    Model Accuracy Note — Updated May 2026

    Current flagship: Claude Opus 4.7 (claude-opus-4-7). Current models: Opus 4.7 · Sonnet 4.6 · Haiku 4.5. Claude Opus 4.7 (claude-opus-4-7) is the current flagship as of April 16, 2026. Where this article references Opus 4.6 or earlier models, those references are historical. See current model tracker →. See current model tracker →

    Claude AI · Fitted Claude

    The short answer is: it depends on what you mean by “smarter.” Claude and ChatGPT are both frontier AI models that perform at similar capability levels on most tasks. Where they differ is in specific strengths, how they handle uncertainty, and the kind of outputs they produce. Here’s the honest breakdown.

    Bottom line: Claude and ChatGPT (GPT-4o) are competitive on most benchmarks. Claude tends to win on writing quality, instruction-following, and honesty calibration. ChatGPT tends to win on ecosystem breadth and image generation. Neither is definitively “smarter” — they have different strengths for different tasks.

    Benchmark Comparison

    Diagram comparing a long context window bar with a shorter output limit bar
    Benchmark comparison.
    Capability Claude Sonnet 4.6 GPT-4o (ChatGPT) Edge
    Writing quality ✅ Stronger Good Claude
    Instruction-following ✅ Stronger Good Claude
    Coding (SWE-bench) ✅ Competitive ✅ Competitive Roughly tied
    Math reasoning ✅ Strong ✅ Strong Roughly tied
    Expressing uncertainty honestly ✅ Stronger More confident Claude
    Context window 1M tokens 128K tokens Claude
    Image generation ❌ Not included ✅ DALL-E built in ChatGPT
    Data analysis (code interpreter) Limited ✅ Advanced Data Analysis ChatGPT
    Hallucination rate ✅ Lower Higher Claude

    Where Claude Is Genuinely Stronger

    Three stacked layers: chat UI, tools, agent runtime
    Where Claude is genuinely stronger.

    Writing quality. Claude produces prose that reads more naturally and holds style constraints more consistently. ChatGPT has recognizable output patterns — a cadence and structure that appears even when you try to tune it away. Claude’s writing is harder to fingerprint as AI-generated.

    Following complex instructions. Give both models a detailed, multi-constraint brief and Claude holds all the constraints through a long response more reliably. ChatGPT tends to gradually drift from earlier constraints as output length increases.

    Honesty about uncertainty. Claude is more likely to say “I’m not sure about this” or “you should verify this” rather than confidently asserting something it doesn’t actually know. This is a calibration advantage — confident wrong answers from ChatGPT have frustrated many users who then don’t catch the error.

    Long-context work. At 1M tokens vs ChatGPT’s 128K, Claude can process significantly more content in a single session — entire codebases, large document stacks, extended research contexts.

    Where ChatGPT Is Genuinely Stronger

    Image generation. DALL-E 3 is built into ChatGPT. Claude doesn’t generate images natively in the web interface. For visual workflows this is a real functional gap.

    Code interpreter. ChatGPT’s Advanced Data Analysis runs Python in the conversation — upload a spreadsheet and get charts, analysis, and interactive data work in the same window. Claude can write code but doesn’t execute it in-chat.

    Ecosystem breadth. OpenAI’s longer history means more third-party integrations, a larger community of people sharing GPT prompts, and more specialized GPTs in the store.

    The Practical Answer

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    The practical answer.

    For text-based professional work — writing, analysis, research, coding, strategy — most users find Claude to be the stronger daily driver. For visual content creation, data analysis in-chat, or workflows built around the OpenAI ecosystem, ChatGPT holds meaningful advantages. Many professionals run both and reach for whichever fits the specific task.

    For the full comparison including pricing, see Claude vs ChatGPT: The Honest 2026 Comparison and Claude Pro vs ChatGPT Plus: Same Price, Different Strengths.

    Frequently Asked Questions

    Is Claude smarter than ChatGPT?

    On writing quality, instruction-following, and honesty calibration — yes. On image generation and interactive data analysis — no. Both are competitive on reasoning and coding benchmarks. Neither is definitively smarter overall; they have different strengths for different task types.

    Is Claude better than GPT-4?

    Claude Sonnet 4.6 and Opus 4.6 compare to GPT-4o (the current GPT-4 model) — not the older GPT-4 Turbo. On most head-to-head comparisons, they’re competitive with Claude holding edges in writing quality and context length, and ChatGPT holding edges in image generation and data analysis tools.

    Should I use Claude or ChatGPT?

    Use Claude as your primary tool if your work is primarily text-based — writing, analysis, coding, research. Use ChatGPT if image generation or in-chat Python execution is central to your workflow. Many professionals use both, with Claude as the daily driver and ChatGPT for its specific capabilities.

    Need this set up for your team?
    Talk to Will →

  • Claude File Upload Size Limits & Context Window (2026)

    Claude File Upload Size Limits & Context Window (2026)

    Last refreshed: May 15, 2026

    Official links: Upload limits help · Try Claude (claude.ai) · Plans & pricing

    Model Accuracy Note — Updated May 2026

    Current flagship: Claude Opus 4.8 (claude-opus-4-8). Current models: Fable 5 · Opus 4.8 · Sonnet 5 · Haiku 4.5. Claude Fable 5 (claude-fable-5) is the current flagship, and Claude Opus 4.8 (claude-opus-4-8) is the current Opus tier, as of June 2026. Where this article references Opus 4.6 or earlier models, those references are historical. See current model tracker →. See current model tracker →

    Direct Answer (August 2026): Claude.ai supports uploads up to 30MB per file and up to 5 files per message (PDFs, DOCX, CSV, TXT, images). Text documents are parsed and converted to tokens within Claude’s 200,000-token context window. For larger datasets exceeding 30MB, developers use the Anthropic API with prompt caching or local embeddings to query multi-gigabyte document repositories.

    Claude AI · Fitted Claude

    Claude supports file uploads in claude.ai and via the API, with specific limits on file size, page count, and number of files. Here are the exact limits for PDFs, images, and other document types, plus what to do when your file is too large.

    Claude File Upload Limits (April 2026)

    File type Max file size Page / length limit Notes
    PDF 32 MB 100 pages Text layer required for reading. Image-only scans need OCR first.
    Images (JPG, PNG, GIF, WebP) 5 MB per image Up to 20 images per request All current Claude models support image input.
    Text files (TXT, MD, CSV) ~10 MB Context window limit Limited by context window, not file size.
    Word / DOCX ~10 MB Context window limit Claude extracts text content.
    Code files Context window limit — No special limit beyond context window.

    What Happens When a File Is Too Large

    Side-by-side of oversized uploads versus split compress link fixes
    Too big? Split, compress, or link — don’t force the upload.

    If a PDF exceeds 32 MB or 100 pages, Claude.ai will reject the upload with an error. The file won’t be processed. The practical workarounds:

    • Split the PDF. Most PDF readers and tools (Preview on Mac, Adobe, Smallpdf) can split a document into smaller sections. Upload the relevant section rather than the full document.
    • Compress the file. Large PDFs are often oversized due to embedded images. Use a PDF compressor to reduce file size while preserving text quality.
    • Copy and paste the text. For text-heavy documents, copying relevant sections directly into the chat removes the file size constraint entirely — the only limit is the context window (1M tokens for Sonnet and Opus).
    • Use multiple conversations. Process different sections in separate conversations and synthesize results yourself.

    Context Window as the True Limit

    Two bars showing upload ceiling versus what actually fits in context
    Context window is the true limit — not just the upload meter.

    Even within the file size limits, the real constraint is the context window — how much text Claude can process at once. A 100-page PDF that’s text-heavy may contain 60,000–80,000 tokens. Claude Sonnet 4.6 and Opus 4.6 have a 1 million token context window, so most documents fit comfortably. Claude Haiku 4.5’s 200,000 token window is still large enough for most individual documents.

    Where the context window becomes the binding constraint is when you’re uploading multiple large files simultaneously — several hundred pages of documents combined may approach context limits on Haiku.

    Scanned PDFs: The Hidden Limit

    Small token cubes assembling into short phrase cards on a desk
    Tokens still assemble the doc you uploaded — count before you paste.

    File size and page count are the official limits, but there’s a functional limit that catches many users: scanned PDFs that are image-only have no text layer, so Claude can’t read their content regardless of size. A 5-page scanned document may be effectively unreadable while a 100-page digital PDF works fine. Run scanned documents through OCR software to create a text layer before uploading. See Can Claude Read PDFs? for the full breakdown.

    Image Limits in Detail

    Each image can be up to 5 MB, with a maximum of 20 images per API request. In Claude.ai conversations, you can upload multiple images in a single message. Claude processes images using its vision capability — all current models (Haiku 4.5, Sonnet 4.6, Opus 4.6) support image input including JPG, PNG, GIF, and WebP formats.

    Frequently Asked Questions

    What is the Claude file size limit?

    PDFs: 32 MB and 100 pages maximum. Images: 5 MB per image, up to 20 images per request. Text files and documents: effectively limited by the context window rather than file size. These limits apply to claude.ai and the API.

    What do I do if my PDF is too large for Claude?

    Split the PDF into smaller sections, compress it to reduce file size, or copy and paste the relevant text directly into the conversation. Text pasted directly is only limited by the context window (1M tokens for Sonnet and Opus), not file size limits.

    How many files can I upload to Claude at once?

    Multiple files can be uploaded in a single conversation. The practical limit is the combined text content fitting within Claude’s context window — 1M tokens for Sonnet 4.6 and Opus 4.6, or 200K tokens for Haiku 4.5. For images, the API supports up to 20 per request.

    Need this set up for your team?
    Talk to Will →

    💼 Deploying Claude or AI Infrastructure in Your Business?

    At Tygart Media, we engineer custom Model Context Protocol (MCP) servers, multi-model content pipelines, and AI operational systems. Explore our Claude AI Team Implementation Services or check out our complete Restoration Operations & AI Kit.

  • Claude Token Limit: Context Windows, Output Limits, and What They Mean in Practice

    Claude Token Limit: Context Windows, Output Limits, and What They Mean in Practice

    Last verified: September 26, 2026 (Pacific Time)

    Model Accuracy Note — Updated September 26, 2026

    Current models (September 2026): Fable 5.1 · Opus 5.5 · Sonnet 5 · Haiku 4.5. Current model tracker →

    Direct Answer (September 2026): Claude’s current models — Fable 5.1, Opus 5.5, and Sonnet 5 — share a 1,000,000-token context window (~750,000 words or ~1,500 pages) with up to 128,000 tokens of max output per response. Haiku 4.5 (the fast, cheap tier) has a 200,000-token context window and 64,000-token max output.

    Claude AI · Fitted Claude

    Claude’s token limits depend on which model you’re using and whether you’re on the web interface or the API. Here are the exact numbers — context window, output limits, and what they mean in practice.

    Key distinction: The context window is the total tokens Claude can process in one conversation (input + output combined). The output limit is the maximum tokens in a single response. These are different limits and both matter depending on your use case.

    Claude Token Limits by Model (September 2026)

    Model Context Window Max Output
    Claude Fable 5.1 1,000,000 tokens 128,000 tokens
    Claude Opus 5.5 1,000,000 tokens 128,000 tokens
    Claude Sonnet 5 1,000,000 tokens 128,000 tokens
    Claude Haiku 4.5 200,000 tokens 64,000 tokens

    Context windows are at standard API pricing (no long-context premium). Claude Opus 5 (released July 24, 2026) is superseded by Opus 5.5 (released September 25, 2026) at a lower price ($4/$20 vs $5/$25 per MTok). Fable 5, Opus 4.8, and Sonnet 4.6 are legacy models — use Opus 4.8 for new projects. Outputs above ~16,000 tokens should be streamed to avoid request timeouts. The original Opus 4 and Sonnet 4 were retired on June 15, 2026 — see the deprecation guide. The asynchronous Message Batches API runs at 50% cost and supports larger per-request outputs for document-generation workloads.

    What a Token Is

    Small token cubes assembling into short phrase cards on a desk
    Count tokens when planning long docs — not characters alone.

    A token is roughly 3–4 characters of English text — about 0.75 words. One page of text is approximately 500–700 tokens. A 200-page book is roughly 100,000–140,000 tokens.

    Content Approx. tokens
    1 word ~1.3 tokens
    1 page of text (~500 words) ~650 tokens
    Short novel (80,000 words) ~104,000 tokens
    Full codebase (10,000 lines) ~100,000–200,000 tokens
    1M token context (Sonnet/Opus) ~750,000 words / ~1,500 pages

    Context Window vs. Output Limit

    Diagram comparing a long context window bar with a shorter output limit bar
    Two different ceilings. Hitting one does not mean you hit the other.

    The context window is the total working memory for a session — everything Claude can “see” at once, including the system prompt, all previous messages in the conversation, uploaded files, and Claude’s own prior responses. At 1M tokens, the legacy (still listed) models (Fable 5, Opus 4.8 (legacy — still listed), Sonnet 4.6 (legacy — still listed)) can hold roughly 1,500 pages of text in context simultaneously.

    The output limit is how long Claude’s individual response can be, and it is model-dependent: up to 128,000 tokens on Claude Fable 5.1, Opus 5.5, and Sonnet 5, and 64,000 tokens on Haiku 4.5. Outputs above roughly 16,000 tokens should be streamed to avoid request timeouts.

    Lineup currency (Sept 2026): Current API list (Sept 2026, verified): Sonnet 5 $2/$10, Opus 5.5 $4/$20, Haiku 4.5 $1/$5, Fable 5.1 $10/$50. Legacy (still listed on Anthropic’s card): Opus 4.8 $5/$25, Sonnet 4.6 $3/$15.

    Rate Limits: Separate From Token Limits

    Token limits are per-conversation. Rate limits are per-time-period — how many tokens (and requests) you can send across multiple conversations in a given minute or day. Rate limits scale with your API usage tier. If you’re hitting errors in production that look like limits, check whether you’re hitting the context window, the output limit, or a rate limit — they produce different error codes. For the full rate limit breakdown, see Claude Rate Limits: What They Are and How to Work Around Them.

    What Happens When You Hit the Context Limit

    Chat transcript fading at the edge of a conversation window when context fills up
    Soft edge: summarize, start a fresh thread, or drop older turns.

    In claude.ai conversations, you’ll see a warning when the conversation is approaching the context window. Claude may summarize earlier parts of the conversation to stay within limits. In the API, sending more tokens than the context window allows returns an error. For very long sessions, breaking work into multiple conversations or using prompt caching (which stores static context at a discount) are the standard approaches.

    Frequently Asked Questions

    What is Claude’s token limit?

    Claude Fable 5, Opus 4.8, and Sonnet 4.6 have a 1 million token context window. Claude Haiku 4.5 has a 200,000 token context window. Maximum output per response is model-dependent — 128,000 tokens on Opus 4.8 and Fable 5, 64,000 on Sonnet 4.6 and Haiku 4.5. These are different limits — context window is total working memory, output limit is maximum response length.

    How long can Claude’s responses be?

    The maximum output is model-dependent — up to 128,000 tokens on Claude Opus 4.8 and Fable 5, and 64,000 tokens on Sonnet 4.6 and Haiku 4.5 (roughly 48,000–96,000 words). Outputs above ~16,000 tokens should be streamed. In practice, claude.ai conversations have shorter limits than the raw API.

    How many tokens is a page of text?

    Approximately 650 tokens per page (roughly 500 words). A 200-page document is around 130,000 tokens — well within Claude’s 1M context window for Sonnet and Opus, and within Haiku’s 200K window as well.

    How many tokens can Claude Opus 4.8 handle?

    Claude Opus 4.8 supports a 1 million token context window for input and up to 128,000 tokens for output. This means you can feed in very large documents, entire codebases, or long conversation histories in a single request.

    What happens when you hit Claude’s context window limit?

    When your input exceeds Claude’s context window, the API returns an error and the request fails. In claude.ai, the interface warns you before sending. The fix is to summarize earlier conversation turns, split the input into chunks, or use a model with a larger context window.

    Is Claude Haiku 4.5 context window smaller than Opus?

    Yes. Claude Haiku 4.5 has a 200,000 token context window, compared to 1 million tokens for Opus 4.8 and Sonnet 4.6. For most tasks Haiku’s 200K window is sufficient, but for very long documents or large codebases, Sonnet 4.6 or Opus 4.8 is required.

    Need this set up for your team?
    Talk to Will →

    💼 Deploying Claude or AI Infrastructure in Your Business?

    At Tygart Media, we engineer custom Model Context Protocol (MCP) servers, multi-model content pipelines, and AI operational systems. Explore our Claude AI Team Implementation Services or check out our complete Restoration Operations & AI Kit.

    >Part of the complete guide: Claude Pricing, Plans & Limits

  • Does Claude Hallucinate? Accuracy & Truthfulness (2026)

    Does Claude Hallucinate? Accuracy & Truthfulness (2026)

    Last refreshed: June 9, 2026

    Claude AI · Fitted Claude

    Yes — Claude hallucinates. Every large language model does. The more useful question is: how often, on what types of tasks, and how does it compare to alternatives? Here’s an honest assessment of where Claude’s hallucination problem is real, where it’s overblown, and how to work with Claude in ways that minimize inaccurate outputs.

    Bottom line: Claude hallucinates less than most alternatives on most benchmarks, and is more likely to express uncertainty rather than confabulate confidently. But hallucination is not eliminated — and Claude is not a reliable source for specific facts, citations, statistics, or recent events without verification.

    Claude Hallucination Risk by Task Type (2026 Assessment)

    Task Type Hallucination Risk Notes
    Factual Q&A (well-known facts) Low High accuracy on stable, widely-documented facts
    Recent events (post-training cutoff) High Claude’s knowledge has a cutoff — it will guess if not told
    Citations & references High Claude may generate plausible-sounding but wrong citations
    Code generation Medium Logic errors more common than syntax errors; always test
    Mathematical calculations Medium Simple arithmetic: good. Multi-step proofs: verify externally
    Summarizing provided documents Low When source material is in context, accuracy is much higher
    Legal / medical specifics Medium-High Use as a starting point only; verify with professionals

    What Hallucination Actually Means

    Three stacked layers: chat UI, tools, agent runtime
    What hallucination actually means.

    Hallucination in AI models means generating plausible-sounding but factually incorrect content. This ranges from subtle errors — slightly wrong dates, invented quotes attributed to real people — to confident fabrications of sources, studies, or events that don’t exist. The model isn’t lying; it’s producing statistically probable text that happens to be wrong.

    Where Claude Hallucinates Most

    Diagram comparing a long context window bar with a shorter output limit bar
    Where Claude hallucinates most.

    Specific citations and sources. Ask Claude to cite a paper, book, or article and it may generate a plausible-looking citation that doesn’t exist — correct author names, plausible journal, wrong or invented title. This is one of the most reliable hallucination triggers across all LLMs, Claude included.

    Statistics and precise numbers. “What percentage of…” questions invite fabrication. Claude will often produce a number that sounds reasonable but has no verified source. When Claude says “studies show X%,” that number may be invented.

    Recent events. Claude’s knowledge has a cutoff date. For events after that date it either refuses to answer, hedges appropriately, or — in the worst case — confabulates based on patterns from its training data.

    Obscure specifics. The more niche the subject, the thinner the training data, and the higher the risk of plausible but wrong outputs. Popular topics have more training data reinforcing correct facts; obscure topics have less.

    Where Claude Is More Reliable

    Reasoning and logic. Claude is significantly better at catching its own errors in structured reasoning than it is at factual recall. Chain-of-thought tasks, mathematical reasoning, and logical analysis are areas where hallucination is less common.

    Expressing uncertainty. One of Claude’s distinctive characteristics is that it’s more likely to say “I’m not certain about this” or “you should verify this” than to confidently assert something it’s unsure about. This calibration is better than most alternatives — though not perfect.

    Well-documented topics. For widely-covered subjects with extensive training data, Claude’s factual accuracy is significantly better than for obscure ones. General knowledge, established science, and well-documented history have lower hallucination rates.

    Claude vs ChatGPT on Hallucination

    On most independent benchmarks, Claude hallucinates at a lower rate than GPT-4o and earlier ChatGPT models. The gap is most noticeable on citation accuracy and on resisting confident confabulation — Claude is more likely to hedge, while ChatGPT has historically been more likely to produce confident wrong answers. The practical difference in everyday use is meaningful but not night-and-day: both models hallucinate on the same types of tasks.

    How to Minimize Hallucination When Using Claude

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    How to minimize hallucination when using Claude.

    Always verify facts independently. Never trust a specific statistic, citation, date, or proper noun from Claude without checking a primary source.

    Ask Claude to flag uncertainty. Add to your prompt: “If you’re not certain about something, say so.” Claude is more reliable when explicitly asked to express uncertainty.

    Don’t ask for citations from memory. Instead, give Claude the source and ask it to work with what you’ve provided. Or use Claude with web search enabled to pull live information.

    Use Claude for reasoning, not recall. The strongest use of Claude is reasoning about information you’ve provided, not retrieving facts from its training data.

    Enable web search for current facts. Claude.ai’s web search integration significantly reduces hallucination on current events and recent data by grounding responses in retrieved content.

    Frequently Asked Questions

    Does Claude hallucinate?

    Yes. Like all large language models, Claude produces factually incorrect content on some portion of responses. It hallucinates most on citations, specific statistics, and obscure topics. It hallucinates less on well-documented subjects and is more likely to express uncertainty than to confabulate confidently.

    Is Claude more accurate than ChatGPT?

    On most benchmarks, yes — Claude hallucinates at a lower rate and is better calibrated to express uncertainty when it doesn’t know something. The practical difference is meaningful but both models have significant hallucination rates on citations and specific facts. Neither should be trusted as a sole source for factual claims.

    How do I stop Claude from hallucinating?

    You can’t eliminate hallucination entirely, but you can minimize it. Provide your own sources rather than asking Claude to recall them. Enable web search for current facts. Ask Claude to flag uncertainty in its responses. Use Claude for reasoning about information you’ve provided rather than as a fact database. Always verify specific claims independently before using them.

    ⬡

    Deploying Claude for your organization?

    We configure Claude correctly — right plan tier, right data handling, right system prompts, real team onboarding. Done for you, not described for you.

    Learn about our implementation service →

    How do I reduce Claude hallucinations in my prompts?

    The most effective techniques: (1) Provide source documents in context rather than asking Claude to recall facts from memory. (2) Ask Claude to cite specific passages from the documents you provide. (3) Instruct Claude to say I don’t know rather than guess. (4) For critical outputs, prompt Claude to review its own answer for factual errors before finalizing.

    Does Claude hallucinate less than GPT or Gemini?

    Benchmark results vary by task and version. Claude generally scores well on factual consistency benchmarks, particularly when source documents are provided in context. However, no current LLM is hallucination-free. Independent evaluations show all frontier models hallucinate on a meaningful percentage of questions, especially on recent or obscure topics.

    Can I trust Claude for medical or legal advice?

    No. Claude is not a licensed medical professional or attorney. It can provide general information, but should not replace professional advice for any medical diagnosis, treatment decision, or legal strategy. Always verify with a qualified professional.

    Need this set up for your team?
    Talk to Will →

  • Claude AI Safety, System Prompts & Guardrails (2026)

    Claude AI Safety, System Prompts & Guardrails (2026)

    Last refreshed: May 15, 2026

    Model Accuracy Note — Updated May 2026

    Current flagship: Claude Opus 4.7 (claude-opus-4-7). Current models: Opus 4.7 · Sonnet 4.6 · Haiku 4.5. Claude Opus 4.7 (claude-opus-4-7) is the current flagship as of April 16, 2026. Where this article references Opus 4.6 or earlier models, those references are historical. See current model tracker →. See current model tracker →

    Claude AI · Fitted Claude

    A Claude jailbreak is any technique designed to bypass Claude’s safety training and get it to produce content it would otherwise refuse. People search for this for different reasons — curiosity about how AI safety works, security research, or genuine attempts to exploit the model. Here’s what jailbreaking Claude actually looks like, why it’s harder than most people expect, and what happens when it does work.

    The honest framing: Claude is the most safety-hardened commercial AI model available in 2026. Standard jailbreak techniques have low single-digit success rates against it. That said, no model is unbreakable — persistent, multi-turn adversarial prompting has demonstrated real-world success. Anthropic publishes its research on this openly and updates defenses continuously.

    How Claude’s Safety System Works

    Three stacked layers: chat UI, tools, agent runtime
    How Claude’s safety system works.

    Claude’s safety isn’t a single content filter — it’s a layered defense built into the model at training time. Anthropic uses Constitutional AI, a technique where Claude is trained against a set of principles and learns to evaluate its own outputs. The model doesn’t just pattern-match on blocked keywords; it reasons about whether a response would cause harm given the full context of the request.

    On top of the trained model, Anthropic adds Constitutional Classifiers — a second layer that monitors inputs and outputs independently, trained on synthetic adversarial prompts across thousands of variations. Compared to an unguarded model, Constitutional Classifiers reduced the jailbreak success rate from 86% to 4.4% — blocking 95% of attacks that would otherwise bypass Claude’s built-in safety training.

    Common Jailbreak Techniques and Why They Don’t Work Well on Claude

    Persona injection (“DAN” / “do anything now”). Asking Claude to adopt an unrestricted persona — an “unfiltered AI,” a fictional character not bound by guidelines. Claude’s Constitutional AI training is robust against most direct persona injection attempts: the model declines the underlying request rather than complying through the fictional wrapper.

    Roleplay framing. Wrapping harmful requests in fictional or hypothetical scenarios — “write a story where a character explains how to…” Claude evaluates the real-world impact of its outputs, not just the fictional framing. A response that would cause harm outside fiction causes the same harm inside it.

    Token manipulation. Base64 encoding, unusual capitalization, Unicode substitution, and other character-level tricks to route requests past classifiers. Constitutional Classifiers are trained on these variations and handle most of them.

    Reasoning framing. Presenting harmful requests as academic, research, or security-related. Claude considers whether a request is plausibly legitimate given context — a genuine security research context differs from a claim of being a researcher with no supporting context.

    Where Jailbreaks Do Work

    Diagram comparing a long context window bar with a shorter output limit bar
    Where jailbreaks do work.

    The Mexico breach in early 2026 — where an attacker used over 1,000 Spanish-language prompts, role-playing Claude as an “elite hacker” in a fictional bug bounty program, eventually causing Claude to abandon its alignment context — demonstrated that persistent multi-turn escalation can work against even hardened models. The attack succeeded not through a clever single prompt but through sustained pressure, context manipulation, and gradual escalation across a long session.

    Multi-turn escalation still works at a non-trivial rate. Single-prompt jailbreaks are mostly defeated. Long sessions with gradual escalation remain a real vulnerability. Anthropic updated Claude Opus 4.6 with real-time misuse detection following the incident.

    Anthropic’s Public Red-Teaming Program

    Anthropic doesn’t just build defenses — it tests them publicly. Over 180 security researchers spent more than 3,000 hours over two months trying to jailbreak Claude using Constitutional Classifiers, offering a $15,000 bounty for a successful universal jailbreak. They weren’t able to find one during that period, though subsequent research has found partial techniques.

    This transparency is part of Anthropic’s approach: publish the research, run public bug bounties, and update defenses based on what adversaries discover. The Constitutional Classifiers paper is publicly available and describes the methodology in full.

    What Happens When Claude Gets Jailbroken

    The consequences range from producing harmful content (the worst case) to simply generating off-policy responses that violate Anthropic’s usage terms. Accounts used to jailbreak Claude are banned. In the Mexico case, Anthropic banned the implicated accounts and shipped defensive updates to the model within weeks of discovery.

    Using jailbreaks to extract harmful content violates Anthropic’s terms of service regardless of intent. Using jailbroken Claude to cause real-world harm — as in the Mexico case — is a criminal matter.

    The Practical Alternative to Jailbreaking

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    The practical alternative to jailbreaking.

    Most people searching for jailbreaks actually want Claude to do something specific it’s currently refusing. Claude’s refusals are mostly a context problem, not a censorship problem. Providing more context about your role, purpose, and authorization frequently resolves apparent refusals that feel like hard limits. If you’re building a product that needs capabilities beyond what the consumer interface allows, the Claude API with appropriate operator system prompts is the legitimate path — not jailbreaking.

    For Claude’s full privacy and safety stance, see Is Claude Safe to Use? and Claude Privacy: What Anthropic Does With Your Data.

    Frequently Asked Questions

    Can Claude be jailbroken?

    Yes, but with difficulty. Standard single-prompt jailbreak techniques have very low success rates against Claude’s Constitutional AI training and Constitutional Classifiers. Persistent multi-turn escalation over long sessions has demonstrated real-world success. Anthropic continuously updates defenses and bans accounts used for jailbreaking.

    Is jailbreaking Claude illegal?

    Jailbreaking violates Anthropic’s terms of service. Using jailbreak techniques to cause real-world harm — breaching systems, generating CSAM, synthesizing weapons — is illegal regardless of the AI tool involved. Anthropic bans accounts and cooperates with law enforcement when illegal activity is discovered.

    Why does Claude refuse some requests that seem harmless?

    Claude evaluates requests as policies — imagining many different people making the same request and calibrating its response to the realistic distribution of intent. Some requests that are genuinely harmless get caught by this calibration. Providing more context about your specific purpose and role usually resolves these cases without needing to “jailbreak” anything.

    ⬡

    Deploying Claude for your organization?

    We configure Claude correctly — right plan tier, right data handling, right system prompts, real team onboarding. Done for you, not described for you.

    Learn about our implementation service →

    Need this set up for your team?
    Talk to Will →

  • Anthropic Console: Login, API Keys & Billing (2026)

    Anthropic Console: Login, API Keys & Billing (2026)

    Last verified: 26 September 2026 (Pacific Time). Console URL move confirmed live: console.anthropic.com now redirects to platform.claude.com.

    Direct Answer (26 September 2026): The Anthropic Console — where you create API keys, buy credits, and monitor usage — now lives at platform.claude.com (the old console.anthropic.com address redirects there). To get an API key: 1) sign up with email, Google, or SSO, 2) add billing under Settings → Plans & Billing (deposit $5+ for Tier 1 activation), 3) open API Keys → Create Key, and 4) copy the sk-ant- key immediately — it is shown exactly once. There is no permanent free API tier; usage is billed per million tokens.

    The console moved: platform.claude.com

    If your bookmarks still say console.anthropic.com, they still work — that address now redirects to platform.claude.com, the current developer console. Older tutorials treat them as different sites; in 2026 they land in the same place. (Also searched as anthopic / anthropic concole — same console, same API keys.)

    What the console is for: API keys, prepaid credit balance and billing, per-key usage monitoring, workspaces and team seats, rate-limit tiers, and the Workbench for testing prompts. claude.ai is the chat product. The console is keys, credits, and API management. The “Anthropic Developer Console” name you see in some docs is the same dashboard, not a second product.

    The console itself is free — there is no monthly fee for access. You only pay for API usage against your credit balance.

    Logging in: console.anthropic.com login goes here now

    Sign in at platform.claude.com with an email code, Google, or SSO. There is no Microsoft or Apple login on the console. New accounts answer brief onboarding questions (individual vs. organization), then land in the dashboard.

    Create an API key — 4 steps

    1. Create your account. Open platform.claude.com and sign up with email, Google, or SSO.
    2. Add billing. Go to Settings → Plans & Billing and add a payment method, then deposit $5+ for Tier 1 activation. Skipping this is the #1 reason a brand-new key returns errors — the key exists, but requests are rejected until the account can be billed.
    3. Create the key. Click API Keys in the left sidebar (direct: platform.claude.com/settings/keys), then Create Key. Name it per environment (my-app-dev, laptop-2026-q3) — future you will thank present you at rotation time. Keys are scoped to a workspace: a key only sees resources in the workspace where it was created.
    4. Copy it immediately. The key is displayed exactly once. It starts with sk-ant- (current keys look like sk-ant-api03-…) and runs roughly 100 characters. Copy it straight into a password manager or secrets manager. Lost it? There is no way to view it again — revoke it and create a replacement (takes a minute, harms nothing).

    Two extras worth setting at creation: a per-key spend cap (no cap on a shared key is how a loop empties a workspace overnight), and a smoke test on Haiku 4.5 — the cheapest model — before pointing real traffic at the key.

    Where to put the key (and where never to put it)

    macOS / Linux:

    export ANTHROPIC_API_KEY="sk-ant-YOUR_KEY_HERE"

    Windows (PowerShell):

    setx ANTHROPIC_API_KEY "sk-ant-YOUR_KEY_HERE"

    Replace YOUR_KEY_HERE with the key shown once at creation. Never commit a key to version control, never paste one into client-side JavaScript (always proxy through a backend), never paste one into a shared chat or doc, and never log full HTTP headers that carry it. GitHub scans for known credential patterns and notifies Anthropic — a committed key gets auto-revoked within minutes. Treat your API key like a credit card number, because that is what it functions as.

    The 2026 no-static-key option: for local development you can skip stored keys entirely with ant auth login (browser OAuth flow), and for CI/production use Workload Identity Federation instead of long-lived secrets.

    Key types and format

    • Live keys (sk-ant-api…) — production inference calls, real billing. Most developers only need this.
    • Admin keys (sk-ant-admin…) — workspace admin operations, not for inference calls.

    If your key does not start with sk-ant-, you copied something incomplete. Use separate keys per environment and per app — so revoking one is surgical instead of catastrophic.

    Billing: the pitfalls that burn credits

    • A chat seat is not console credit. Claude Pro, Max, and Team do not add API credits. API usage draws from a separate prepaid console balance.
    • An empty prepaid balance looks like an API outage. Failed calls with a $0 balance return billing errors — check Billing before debugging code.
    • Your rate-limit tier follows deposited spend, not your chat plan. A new workspace starts tight; raising the prepaid total is what moves the tier.
    • Set auto-recharge and a monthly usage limit. Auto top-ups with a low-balance threshold plus a monthly cap are the cheapest insurance against a runaway loop or a compromised key. Enterprise accounts can invoice instead of prepaying.

    What the API costs (September 2026)

    Per million tokens, input / output:

    Model Input Output
    Haiku 4.5 $1 $5
    Sonnet 5 $2 $10
    Opus 5.5 $4 $20
    Opus 5 $5 $25
    Fable 5.1 $10 $50

    The Batch API cuts those rates 50% for async workloads. Prompt-cache reads run about 10% of base input price. Full seat and API tables live on our Claude pricing hub, re-verified 26 September 2026.

    Security: rotate before you regret it

    Rotate keys quarterly at minimum, immediately if compromise is suspected:

    1. In API Keys, create a new key with a fresh dated name (laptop-2026-q4).
    2. Update the environment variable or secret manager.
    3. Verify the new key works (Haiku smoke test).
    4. Revoke the old key.

    If a key leaks: revoke it in the console immediately, create a replacement, and review the usage logs for unfamiliar spend. The five-minute rotation is far cheaper than dealing with a drained balance.

    Workspaces and team keys

    The hierarchy is Organization → Workspaces → API Keys. Each key belongs to one workspace; billing rolls up to the organization. Need separate billing visibility per project? Separate workspaces are the lever. Invite teammates with specific roles, and set per-user or per-workspace spending limits with alerts.

    Monitoring usage

    The console’s Usage section shows per-key token spend per day, model breakdown, input/output split, and cache usage — with real-time and historical charts by model and time period. Set email alerts on daily/monthly thresholds; it is the cheapest early warning for a runaway loop.

    Rate limits and service tiers

    Limits show in-console as requests/minute and tokens/minute per model, and they grow as the account’s spending history grows. Anthropic’s service tiers are Priority, Standard, and Batch (Batch = async at 50% off) — a separate concept from the Tier 1 activation unlocked by the $5+ billing deposit. New workspaces start with tight limits; deposit history loosens them.

    Troubleshooting the four common errors

    • 401 authentication_error — key missing, mistyped, or revoked. Subtle 2026 variant: if both ANTHROPIC_API_KEY and ANTHROPIC_AUTH_TOKEN are set, the SDK sends both and the API rejects the request — unset one.
    • 403 permission_error — the key works but lacks access to that model or feature; check the key’s workspace and the org’s model access.
    • 429 rate_limit_error — sending faster than the usage tier allows. The response carries a retry-after header; official SDKs retry automatically.
    • Key created but every request fails — almost always incomplete billing (Step 2). Add a payment method and credits, then retry.

    Test your key

    curl https://api.anthropic.com/v1/messages \
      -H "x-api-key: $ANTHROPIC_API_KEY" \
      -H "anthropic-version: 2023-06-01" \
      -H "content-type: application/json" \
      -d '{"model": "claude-haiku-4-5", "max_tokens": 32,
           "messages": [{"role": "user", "content": "Say OK"}]}'

    A JSON response with a content array means you are live.

    FAQ

    Where is the API Keys page in the console?

    Sign in at platform.claude.com, then API Keys in the left nav — or go straight to platform.claude.com/settings/keys → Create Key.

    Where do I find my existing API key?

    You can’t — Anthropic shows keys only at creation. Revoke the old one and create a replacement.

    Is the Anthropic Console free to use?

    Yes — console access has no monthly fee. You pay only for API usage, billed per million tokens against prepaid credits. There is no permanent free API tier.

    Does my Claude Pro or Max subscription include API credits?

    No. Pro/Max are chat and Claude Code seats. API credits are a separate console balance.

    What’s the difference between a Claude API key and a Claude Pro subscription?

    Pro ($20/mo) is the claude.ai web and app chat product. An API key is programmatic access for your own apps and tools — billed separately, per token.

    Why is my first API call failing?

    Usually a missing payment method, a $0 credit balance, or a retired model ID. Fix Billing first, then retry on Haiku 4.5.

    What does an Anthropic API key start with?

    Live keys start with sk-ant-api (current format sk-ant-api03-…), roughly 100 characters. Admin keys start with sk-ant-admin.

    Can I use the same API key for Claude Code and my own app?

    You can, but you shouldn’t. Separate keys per environment make revocation surgical instead of catastrophic. Note: Claude Code (CLI) uses your API key outside a subscription context; Claude Cowork (desktop app) uses your subscription, not a raw key.

    Can I set spending limits on the console?

    Yes — per-workspace and per-user limits, auto top-ups, and email spending alerts, all in Settings → Billing.

    Sources

    I write pages like this so AI search cites them — then I do the same for restoration companies. That’s what Tygart Media does.

  • Claude Code vs Cursor: Which AI Coding Tool is Better?

    Claude Code vs Cursor: Which AI Coding Tool is Better?

    Last refreshed: May 15, 2026

    Claude AI · Fitted Claude

    Claude Code and Cursor are both AI coding tools with serious developer followings — but they’re built on fundamentally different models. Cursor is an AI-powered IDE fork. Claude Code is a terminal-native agent. The right choice depends on how you work.

    Short answer: Cursor wins for in-editor experience — autocomplete, inline suggestions, and staying inside VS Code’s familiar interface. Claude Code wins for autonomous multi-step tasks — it operates at the system level, can run commands, manage files across the whole project, and doesn’t require you to be watching. Most serious developers end up using both.

    Claude Code vs Cursor: Head-to-Head

    Three stacked layers: chat UI, tools, agent runtime
    Claude Code vs Cursor — head-to-head.
    Capability Claude Code Cursor Edge
    In-editor autocomplete Limited ✅ Native Cursor
    Autonomous multi-file tasks ✅ Strong ✅ Good Claude Code
    Terminal / shell command execution ✅ Yes Limited Claude Code
    Remote / cloud sessions ✅ Yes — Claude Code
    VS Code compatibility Via MCP ✅ Built on VS Code Cursor
    Model choice Claude only Multi-model Cursor (flexibility)
    Instruction-following precision ✅ Strong Good Claude Code
    Price Included in Pro ($20/mo)+ ~$20/mo (Pro) Tie
    Setup complexity Moderate Easy Cursor
    Not sure which to use?

    We’ll help you pick the right stack — and set it up.

    Tygart Media evaluates your workflow and configures the right AI tools for your team. No guesswork, no wasted subscriptions.

    What Cursor Does Better

    Long paper tape measure unrolling across a desk beside a laptop, metaphor for context window length
    What Cursor does better.

    In-editor experience. Cursor is a fork of VS Code with AI baked in — autocomplete, inline suggestions, cmd+K to edit code in place, and the full VS Code extension ecosystem. If you live in an editor and want AI suggestions as you type, Cursor is the more polished experience.

    Familiar interface. If your team already uses VS Code, Cursor requires almost no adjustment. Claude Code requires getting comfortable with an agentic workflow that’s fundamentally different from autocomplete.

    Multi-model flexibility. Cursor lets you choose between Claude, GPT-4o, and other models depending on the task. Claude Code is Claude-only.

    What Claude Code Does Better

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    What Claude Code does better.

    System-level autonomy. Claude Code runs commands, manages files across the entire project, executes tests, and operates at the OS level — not just inside an editor window. It can do things Cursor can’t, like run a test suite, see the results, fix the failures, and re-run without you touching anything.

    Remote and background sessions. Claude Code supports remote sessions that continue on Anthropic’s infrastructure even after you close the app. Cursor requires you to be present.

    Complex multi-step tasks. Agentic tasks that span many files, require running code, and iterate based on output are where Claude Code’s architecture shines. Cursor handles this through its Composer feature, but Claude Code’s terminal-native approach gives it more flexibility.

    Instruction precision. On multi-constraint tasks — “refactor this to match our conventions, add error handling, keep it backward compatible, and don’t use async” — Claude Code holds all the constraints more reliably through a long operation.

    Price Comparison

    Claude Code is included (at limited levels) with a Claude Pro subscription at $20/month. Claude Code Pro at $100/month gives full access for developers using it as a primary tool. Cursor Pro is approximately $20/month. Both are in the same price tier for comparable usage levels.

    The Practical Setup

    Most developers using both tools run Cursor for in-editor work — autocomplete, inline edits, quick questions about code — and Claude Code for larger autonomous tasks: refactors, test generation across a codebase, debugging sessions that require running code. They’re complementary, not mutually exclusive.

    For a broader comparison, see Claude vs GitHub Copilot and Claude Code vs Windsurf. For Claude Code pricing specifically, see Claude Code Pricing: Pro vs Max.

    Frequently Asked Questions

    Is Claude Code better than Cursor?

    They’re different tools. Claude Code is better for autonomous multi-step tasks, system-level operations, and complex refactors that require running code and iterating. Cursor is better for in-editor autocomplete and inline suggestions within the VS Code interface. Most serious developers use both.

    Can I use Claude Code inside VS Code or Cursor?

    Claude Code primarily runs as a terminal agent or through Claude Desktop’s Code tab. You can connect it to VS Code via MCP integration. Cursor has its own Claude integration built in — you can use Claude models inside Cursor without Claude Code.

    How much does Cursor cost vs Claude Code?

    Cursor Pro is approximately $20/month. Claude Code is included at limited levels with Claude Pro ($20/month) or at full access with Claude Code Pro ($100/month). For occasional use, Claude Pro gives you both a full Claude subscription and limited Claude Code access for the same $20.

    Need this set up for your team?
    Talk to Will →

    Deeper Dives: Claude Code vs Cursor

  • How to Use Claude Code: Workflows & Best Practices

    How to Use Claude Code: Workflows & Best Practices

    Last refreshed: May 15, 2026

    Model Accuracy Note — Updated May 2026

    Current flagship: Claude Opus 4.7 (claude-opus-4-7). Current models: Opus 4.7 · Sonnet 4.6 · Haiku 4.5. Claude Opus 4.7 (claude-opus-4-7) is the current flagship as of April 16, 2026. Where this article references Opus 4.6 or earlier models, those references are historical. See current model tracker →. See current model tracker →

    Claude AI · Fitted Claude

    Claude Code is an agentic coding tool that works directly inside your codebase — reading files, writing code, running tests, and making changes with your approval. It’s fundamentally different from asking Claude questions about code in chat. Here’s how to use it effectively from your first session.

    Access: Claude Code is in the Code tab of Claude Desktop, or via the CLI (claude command). Both require a Pro, Max, Team, or Enterprise subscription. For installation, see How to Install Claude Code.

    Starting a Session

    Long paper tape measure unrolling across a desk beside a laptop, metaphor for context window length
    Starting a session.

    Open the Code tab in Claude Desktop (or run claude in your terminal), then select your session type:

    • Local — Claude works on your machine with direct file access. Changes happen locally. Best for everyday development.
    • Remote — Claude runs on Anthropic’s infrastructure. The session continues even if you close the app. Best for long autonomous tasks.
    • SSH — Claude connects to a remote server over SSH. Best for server-side or cloud VM development.

    Select your project folder and choose a model — Sonnet 4.6 for most work, Opus 4.6 for the most complex tasks. You can’t change the model after the session starts.

    Free — no pitch

    Get the Claude workflow that actually sticks.

    Practical Claude setup tips from someone running it across 27 client sites daily — not marketing, not theory. Email Will directly and he’ll share what’s working.

    Get the tips →

    What to Ask Claude Code to Do

    Three stacked layers: chat UI, tools, agent runtime
    What to ask Claude Code to do.

    Claude Code works best with clear, specific tasks. Unlike chat, it has your full codebase as context — so you can refer to actual files, functions, and patterns without explaining them.

    Strong prompts:

    • “Fix the bug in auth.py where users aren’t logged out after session expiry”
    • “Add input validation to all API endpoints in routes/ — check for SQL injection”
    • “Refactor UserService to use dependency injection”
    • “Write tests for all public methods in utils/parser.js“
    • “Explain the data flow from the API call in main.py to the database”

    How Claude Code Makes Changes

    Claude Code doesn’t modify your files silently. When it wants to make a change, it shows you a diff — the exact lines it plans to add, remove, or edit — and waits for your approval before writing anything. You can accept the change, reject it, or ask Claude to revise its approach.

    This review-before-write pattern means Claude Code is safe to use even on code you’re not fully familiar with. You see every change before it happens.

    CLAUDE.md: Persistent Project Instructions

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    CLAUDE.md — persistent project instructions.

    Create a CLAUDE.md file in your project root to give Claude Code persistent context about your codebase — coding conventions, architecture decisions, testing patterns, libraries to prefer or avoid. Claude Code reads this file at the start of every session, so you don’t re-explain your project each time.

    # CLAUDE.md example
    
    ## Stack
    - Python 3.11, FastAPI, PostgreSQL, SQLAlchemy
    - Tests use pytest with fixtures in conftest.py
    
    ## Conventions
    - Functions should have type hints
    - No bare except clauses — always catch specific exceptions
    - Use Pydantic models for all request/response schemas
    
    ## Do not modify
    - legacy/ directory — kept for backward compatibility

    Common Workflows

    Debugging: Paste the error message and stack trace, ask Claude to find and fix the root cause. It reads the relevant files and proposes a fix.

    Refactoring: Describe the refactor goal — “extract this into a separate service,” “replace all these conditionals with a strategy pattern” — and Claude Code walks through the changes systematically.

    Test writing: “Write unit tests for all functions in this file” or “add edge case tests for the payment processing flow.” Claude reads your existing test patterns and matches them.

    Code review: “Review this PR for security issues, missing error handling, and anything that violates our conventions.” Claude reads the changed files and gives structured feedback.

    Documentation: “Add docstrings to all public functions” or “generate a README that explains this project’s architecture.”

    Tips for Getting the Most Out of It

    • Start small. First session, give it a small, contained task in a file you know well. Calibrate your trust before handing it larger work.
    • Be specific about constraints. “Don’t use async/await,” “keep changes backward compatible,” “match the style of the existing tests.” Claude Code holds these throughout the session.
    • Use CLAUDE.md. Any convention you find yourself re-explaining belongs in this file.
    • Review every diff. Claude Code is capable but not infallible. Read each proposed change before approving — especially on production code.
    • Use Remote for long tasks. Autonomous multi-file refactors or test generation across a large codebase work better in a Remote session that won’t be interrupted.

    For pricing details, see Claude Code Pricing: Pro vs Max. For how it compares to other tools, see Claude vs GitHub Copilot and Claude Code vs Windsurf.

    Tygart Media

    Getting Claude set up is one thing.
    Getting it working for your team is another.

    We configure Claude Code, system prompts, integrations, and team workflows end-to-end. You get a working setup — not more documentation to read.

    See what we set up →

    Frequently Asked Questions

    How does Claude Code work?

    Claude Code reads your project files, executes tasks like debugging, refactoring, and test writing, and shows you a diff of every proposed change before modifying anything. You approve or reject each change. It operates in your actual development environment, not in a chat window.

    Can Claude Code modify my files without permission?

    No. Claude Code shows you a diff of every proposed change and waits for your approval before writing to any file. You remain in full control of what gets modified.

    What languages does Claude Code support?

    Claude Code works with any programming language. It’s particularly strong on Python, JavaScript, TypeScript, and Go, but handles most modern languages and frameworks well. It reads and writes whatever files are in your project directory.

    Need this set up for your team?
    Talk to Will →

  • Install Claude Code: Complete Desktop & CLI Setup Guide

    Install Claude Code: Complete Desktop & CLI Setup Guide

    Last refreshed: May 15, 2026

    Looking for quick answers? The FAQ version covers every common question directly.

    → Claude Code FAQ

    Quick Answer

    The fastest way to install Claude Code is one npm command: npm install -g @anthropic-ai/claude-code. Full Claude Code installation takes under 2 minutes on macOS, Linux, or Windows (via WSL).

    Full installation walkthrough below — CLI and Desktop app.

    Claude AI · Fitted Claude

    Claude Code is available two ways: as the Code tab inside Claude Desktop (with a graphical interface), or as a CLI tool you install and run from your terminal. Here’s how to get either one set up from scratch.

    Requirement: Claude Code requires a Pro, Max, Team, or Enterprise subscription. It is not available on the free plan. The Claude Desktop app (which includes a graphical Claude Code interface) is free to download but the Code tab requires a paid subscription.

    Option 1: Claude Desktop (Recommended for Most Users)

    Three stacked layers: chat UI, tools, agent runtime
    Option 1 — Claude Desktop.

    The easiest way to get Claude Code is through Claude Desktop — no terminal required.

      Installing Claude?

      If you run into setup issues or want to know how to get the most out of it from day one, just email me.

      Email Will → will@tygartmedia.com

    1. Download Claude Desktop from claude.ai/download — available for macOS and Windows (x64 or ARM64). Linux is not supported.
    2. Install — on Mac, open the PKG and drag to Applications; on Windows, run the installer.
    3. Sign in with your Anthropic account (Pro, Max, Team, or Enterprise).
    4. Click the Code tab in the top navigation.
    5. Select Local to work with files on your machine, or Remote to run on Anthropic’s cloud infrastructure.
    6. Click “Select folder” and choose your project directory. You’re ready.

    On Windows, Git must be installed for local sessions to work. Most Macs include Git by default — check by running git --version in Terminal.

    Free — no pitch

    Get the Claude workflow that actually sticks.

    Practical Claude setup tips from someone running it across 27 client sites daily — not marketing, not theory. Email Will directly and he’ll share what’s working.

    Get the tips →

    Option 2: Claude Code CLI

    Long paper tape measure unrolling across a desk beside a laptop, metaphor for context window length
    Option 2 — Claude Code CLI.

    For developers who prefer working in the terminal, Claude Code is also available as a command-line tool.

    # Install via npm
    npm install -g @anthropic-ai/claude-code
    
    # Authenticate
    claude login
    
    # Start in your project directory
    cd your-project
    claude

    The CLI requires Node.js. After running claude login, you’ll authenticate with your Anthropic account in a browser window. The session starts automatically in the current directory.

    Local vs. Remote Sessions

    Session type What it does Best for
    Local Runs on your machine, accesses your files directly Everyday development work
    Remote Runs on Anthropic’s cloud, continues if you close the app Long-running tasks, autonomous work
    SSH Connects to a remote machine over SSH Server or cloud VM development

    Claude Code Install Errors and How to Fix Them

    Code tab not appearing in Desktop: Confirm your account is on a paid plan. Claude Code requires Pro, Max, Team, or Enterprise — it’s not available on the free tier.

    Windows Git error: Claude Code needs Git for local sessions on Windows. Download Git from git-scm.com, install with default settings, then restart the desktop app.

    CLI authentication failing: Run claude logout then claude login again. Make sure your Anthropic account has an active paid subscription.

    Permission errors on first run: Claude Code will ask permission to access your files when you first select a folder. Click Allow — it needs read/write access to work with your project.

    After Claude Code Installation: Your First Session

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    After installation — your first session.

    When you start your first Claude Code session, Anthropic recommends starting with a small, familiar project. Ask Claude to explain the codebase, fix a specific bug, or add a small feature. This gives you a calibrated sense of how it works before tackling larger tasks. Claude will read relevant files, propose changes, and ask for your approval before modifying anything.

    For an overview of what Claude Code can do once you’re set up, see How to Use Claude Code. For pricing details, see Claude Code Pricing: Pro vs Max.

    Tygart Media

    Getting Claude set up is one thing.
    Getting it working for your team is another.

    We configure Claude Code, system prompts, integrations, and team workflows end-to-end. You get a working setup — not more documentation to read.

    See what we set up →

    Frequently Asked Questions

    How do I install Claude Code?

    Download Claude Desktop from claude.ai/download and use the Code tab — no terminal required. Or install the CLI with npm install -g @anthropic-ai/claude-code and run claude login to authenticate.

    Is Claude Code free to install?

    Claude Desktop (which includes Claude Code) is free to download. Using Claude Code requires a paid subscription — Pro ($20/month), Max ($100/month), Team, or Enterprise. It is not available on the free plan.

    Does Claude Code work on Linux?

    The Claude Desktop app does not support Linux. The Claude Code CLI does run on Linux — install via npm and use it from your terminal.

    What’s the difference between Claude Code Desktop and the CLI?

    Claude Code Desktop (the Code tab in the Claude Desktop app) gives you a graphical interface with visual file diffs, a built-in preview panel, and no terminal required. The CLI runs in your terminal and supports the same core operations. Both share configuration files and can run simultaneously on the same project.

    Need this set up for your team?
    Talk to Will →

    📎 Book for Bots — Free

    Take this article on steroids.

    The Claude Implementation Playbook is a dense 9-section PDF you can attach directly to any AI conversation — pricing tables, model API strings, routing logic, context engineering rules. Verified May 2026.

    Get Free PDF →

    2026 Performance Boost: Higher Limits, No Peak-Hour Throttling

    Added May 9, 2026

    If you installed Claude Code before May 2026 and have noticed it running faster during previously-slow periods, this is why.

    What Changed: SpaceX Colossus 1 Capacity Expansion

    On May 6, 2026, Anthropic announced expanded compute capacity via a partnership with SpaceX’s Colossus 1 data center. The practical effect for Claude Code users:

    • Pro and Max subscribers: Peak-hours throttling eliminated entirely. Claude Code no longer slows down during high-demand periods. 5-hour rate limits doubled.
    • Team Premium seats: 5-hour rate limits doubled. Claude Code access included (Team Premium seats only — Standard seats do not include Claude Code).
    • Free plan: No change. Explicitly excluded from all capacity increases.

    This is the single most meaningful performance change to Claude Code since its GA release. If you’ve been hitting rate limits mid-session during long agentic coding runs, the ceiling is now twice as high. Source: anthropic.com/news/higher-limits-spacex

    Claude Code Pro Pricing (Current)

    Claude Code access is included in the following Claude subscription tiers — there’s no separate Claude Code subscription fee:

    • Pro ($20/month): Claude Code included. 5-hour rate limits now doubled. Peak-hours throttling eliminated.
    • Max ($100/month — 5× usage): Claude Code included. Same doubled limits, no peak throttling.
    • Max ($200/month — 20× usage): Claude Code included. Highest usage ceiling available to individuals.
    • Team Premium ($100/seat/month annually): Claude Code included. 5-hour limits doubled.
    • Team Standard ($25/seat/month annually): Claude Code NOT included.
    • Free plan: Claude Code NOT included.

    Claude Code and Student / Amazon Prime Discounts

    There is no Claude Code-specific discount program and no Amazon Prime Student bundle that includes Claude Code as of May 2026. Claude Pro ($20/month) is the lowest-cost tier that includes Claude Code. For students: Anthropic does not offer a verified student discount on subscriptions. The most cost-effective path to Claude Code for students is the standard Pro plan. See our Claude student discount guide for the full breakdown of what options exist.

    Claude Code Version: Stay Current

    Claude Code releases frequently. As of May 9, 2026, the latest version is v2.1.133 (released May 7). Key fix in this version: subagents now correctly discover project, user, and plugin skills via the Skill tool — a silent failure in prior versions that broke multi-agent pipelines without obvious error. Update with:

    npm update -g @anthropic-ai/claude-code

    >Part of the complete guide: Claude Code

  • Claude Desktop: Features, Download & Web App Differences

    Claude Desktop: Features, Download & Web App Differences

    Last refreshed: May 15, 2026

    Claude AI · Fitted Claude

    Installing Claude?

    If you run into setup issues or want to know how to get the most out of it from day one, just email me.

    Email Will → will@tygartmedia.com

    Claude Desktop is Anthropic’s native app for Mac and Windows — a dedicated application that brings Claude out of the browser and into your system, with features that aren’t available in the web interface. It’s free to download for all plan types, including the free tier.

    Download: claude.ai/download — available for macOS and Windows. Linux is not currently supported. The app is free to download; some features require a paid plan.

    What Claude Desktop Includes

    Three stacked layers: chat UI, tools, agent runtime
    What Claude Desktop includes.
    Feature What it does Plan required
    Chat tab Full Claude conversation, same as claude.ai Free+
    Quick access hotkey Bring Claude up from any app without switching windows Free+
    Screenshot capture Snap your screen and have Claude analyze what it sees Free+
    Desktop extensions Connect Claude to local files, calendars, email, and native apps Free+
    Claude in Chrome Claude navigates, clicks, and fills forms in your browser Free+
    Cowork tab Autonomous background agent that runs tasks in a cloud VM Pro, Max, Team, Enterprise
    Code tab (Claude Code) Interactive coding with direct access to your local files Pro, Max, Team, Enterprise
    Free — no pitch

    Get the Claude workflow that actually sticks.

    Practical Claude setup tips from someone running it across 27 client sites daily — not marketing, not theory. Email Will directly and he’ll share what’s working.

    Get the tips →

    The Three Tabs: Chat, Cowork, and Code

    Side-by-side cards defining what Claude Code is and is not
    The three tabs — Chat, Cowork, and Code.

    Claude Desktop is organized around three modes. Chat is general conversation — identical to claude.ai but accessible from your dock without opening a browser. Cowork is an autonomous background agent that works on tasks in a cloud VM while you do other things — research synthesis, file organization, document generation, spreadsheets. Code is Claude Code with a graphical interface — an interactive coding assistant with direct access to your local project files, where you review and approve each change in real time.

    What Makes Desktop Different From the Web

    The practical differences over using claude.ai in a browser:

    • Always accessible — lives in your dock, reachable from any app via hotkey without switching contexts
    • Screenshot capture — take a screenshot and immediately ask Claude about what’s on screen
    • Desktop extensions — connect to local files, your filesystem, calendars, email, and native Mac/Windows apps that web connectors can’t reach
    • Claude in Chrome — let Claude control your browser directly from the desktop app
    • Cowork background tasks — long-running autonomous tasks that continue while you work on something else
    • Claude Code integration — work directly with your local codebase without copying files

    How to Download and Install

    Desk with laptop, checklist notebook, and billing card ready before creating an Anthropic API key
    How to download and install.
    1. Go to claude.ai/download
    2. Download the appropriate installer — macOS PKG, Windows x64, or Windows ARM64
    3. Run the installer — on Mac, drag to Applications; on Windows, run the setup file
    4. Launch from your Applications folder (Mac) or Start menu (Windows)
    5. Sign in with your Anthropic account

    Linux is not currently supported. The app requires macOS or Windows.

    Desktop Extensions

    Desktop extensions are installable packages that connect Claude to your local tools — similar to browser extensions but for native applications. They give Claude access to your local file system, calendars, email, messaging apps, and other native applications that web-based connectors can’t reach. Browse and install extensions in Settings → Extensions after installing the app.

    Syncing Across Devices

    Your conversations, projects, memory, and preferences sync across Claude Desktop, the web interface, and the mobile app when you’re signed in with the same account. Start a session in the desktop app, continue on your phone, pick it back up on the web — the context travels with you.

    For the Claude Code-specific features in the desktop app, see How to Use Claude Code. For a comparison of Desktop vs. the web interface for everyday use, see Claude AI Login: All Platforms and Apps.

    Tygart Media

    Getting Claude set up is one thing.
    Getting it working for your team is another.

    We configure Claude Code, system prompts, integrations, and team workflows end-to-end. You get a working setup — not more documentation to read.

    See what we set up →

    Frequently Asked Questions

    Is Claude Desktop free?

    Yes. Claude Desktop is free to download and use with a free Claude account. Some features — Cowork, Claude Code, and some extensions — require a paid plan (Pro, Max, Team, or Enterprise). The core chat functionality is available on the free tier.

    What’s the difference between Claude Desktop and claude.ai?

    Claude Desktop is a native app that lives in your dock and adds features not available in the browser — quick access hotkey, screenshot capture, desktop extensions, Cowork for background tasks, and Claude Code for local development. The core conversation experience is the same; the desktop app adds system-level integration.

    Is Claude Desktop available on Linux?

    No. Claude Desktop currently supports macOS and Windows only. Linux users can access Claude through the web interface at claude.ai or via the API.

    Does Claude Desktop work with Claude Code?

    Yes. Claude Desktop includes a Code tab that gives you Claude Code with a graphical interface — direct access to your local files, code review, and interactive development without needing the terminal. The desktop app and the Claude Code CLI share configuration and can run simultaneously on the same project.

    Need this set up for your team?
    Talk to Will →