Category: Claude AI

Extended thinking is Claude’s most powerful reasoning mode — and the one most people never use correctly. This guide explains what extended thinking does, when it genuinely improves outputs, how to enable it, and when you’re better off with a standard prompt.

What Is Extended Thinking?

Extended thinking gives Claude a dedicated reasoning phase before generating its final response. Claude works through a problem on “scratch paper” before writing its answer — exploring multiple approaches, identifying errors in its own reasoning, and building a more deliberate chain of thought. On current Claude models (Fable 5, Opus 4.8, Sonnet 4.6), this works through adaptive thinking — Claude dynamically decides when and how much to think based on problem complexity, instead of you setting a fixed token budget.

When Extended Thinking Genuinely Helps

• Complex math and logic problems requiring step-by-step reasoning
• Multi-step coding tasks with many interdependent components
• Strategic analysis requiring weighing many variables
• Difficult research synthesis where accuracy matters most
• Any task where “think step by step” would help — extended thinking does this automatically

When Extended Thinking Is Overkill

• Simple factual questions with clear answers
• Routine writing tasks (emails, summaries, short copy)
• Format conversion or data transformation
• Tasks where speed matters more than depth

How to Enable Extended Thinking

In Claude.ai: Look for the thinking toggle before sending your message.

Via API (current models — Fable 5, Opus 4.8, Sonnet 4.6): Use adaptive thinking and let Claude decide the depth:

"thinking": { "type": "adaptive" },
"output_config": { "effort": "high" }

The effort parameter — low, medium, high, xhigh, or max — controls how much Claude thinks and how many tokens it spends. Default is high. Use xhigh for coding and agentic work; low for fast, simple tasks. (xhigh is available on Fable 5 and Opus 4.7+; max on Opus-tier 4.6+ and Sonnet 4.6.)

What You See During Extended Thinking

In claude.ai, Claude shows a collapsed “thinking” section before its response. Expand it to see the reasoning chain — useful for verifying logic or understanding how Claude approached a problem. The thinking section is exploratory and may contain dead ends; this is normal.

Via API, on Fable 5, Opus 4.7, and Opus 4.8 the thinking text is omitted by default (the blocks stream but their text is empty). To see summarized reasoning, request "thinking": {"type": "adaptive", "display": "summarized"}.

Related on Tygart Media: how to use Claude · Claude Code tutorial · is Claude worth it.

Frequently Asked Questions

Does extended thinking always give better answers?

No. It improves accuracy on complex reasoning tasks but adds latency. For simple tasks, standard mode is faster and just as accurate.

Claude’s memory feature changes the product from a stateless chatbot into something that actually knows you. Without memory, Claude starts from zero every conversation. With memory configured, Claude builds a growing knowledge base about you that it draws on automatically. This guide explains how it works and how to get the most from it.

How Claude Memory Works

Claude’s memory is an auto-synthesized knowledge base. Approximately every 24 hours, the system reviews recent conversations and extracts facts, preferences, and patterns worth remembering — then stores those as structured memory entries. Memory is separate for Projects vs. standalone conversations — each Project has its own memory space.

What Claude Can Remember

• Your name, role, and professional context
• Preferred communication style and tone
• Ongoing projects and their context
• Tools, frameworks, and workflows you use
• Output format preferences
• Things you’ve asked Claude not to do

How to Configure Memory

In Claude.ai, go to Settings → Memory. You’ll see auto-generated memory entries. You can review, edit, delete, or manually add memories. You can also instruct Claude directly: “Remember that I prefer bullet points” or “Don’t forget my target audience is non-technical executives.”

Memory vs. Project Instructions

Project instructions are static — written once, apply to every conversation. Memory is dynamic — evolves as Claude learns. Use Project instructions for consistent role context. Use memory for personal preferences and evolving project context.

CLAUDE.md for Claude Code

For Claude Code, place a CLAUDE.md file in your project root. Claude Code reads it at the start of every coding session. Use it for: project architecture, coding standards, common patterns, known issues. This is the most powerful memory tool for developers.

Frequently Asked Questions

Does Claude remember everything I say?

No. Memory synthesizes and stores key facts and preferences, not verbatim conversation logs. It’s selective — designed to capture what’s useful.

Can I delete Claude’s memories about me?

Yes. Go to Settings → Memory in Claude.ai to view and delete any memory entries.

Direct answer: Claude does not generate photos or illustrations the way Midjourney, DALL·E, or Gemini Imagen do. It can analyze images you upload, build diagrams and interactive visuals with HTML and SVG in chat (beta on web and desktop), and—through MCP connectors—call external image models when you wire them up. For operators tracking how AI answers cite your brand, that split matters when you document tools in content and monitor AI citations.

What Claude still cannot do: native image generation

Anthropic’s help center states plainly that Claude doesn’t generate photos or illustrations like dedicated image tools (Can Claude produce images?, updated March 2026). That has not changed across current Claude model releases in 2026. If you need marketing hero shots, photorealistic scenes, or meme-style art, use a purpose-built generator:

• ChatGPT — built-in GPT Image / DALL·E workflow inside one app
• Midjourney — strong aesthetic control for creative and brand imagery
• Adobe Firefly — commercial-friendly generation inside Creative Cloud
• Google Gemini — Imagen-family output in the Gemini app
• Stable Diffusion — open weights; self-host or use a hosted wrapper

Claude’s value is elsewhere: reading job-site photos, summarizing damage docs, and drafting the copy that goes beside those assets—work that shows up when you measure whether AI surfaces your pages in LLM visibility programs.

What Claude can do with images and visuals

Vision: analyze and describe uploads

Upload photos, screenshots, PDF scans, or charts. Claude describes content, spots damage or equipment, reads labels, and answers targeted questions. That includes handwriting and cramped table screenshots—common on restoration estimates and insurer portals.

Text extraction (OCR-style)

Point Claude at a whiteboard photo, moisture log, or phone picture of a printed scope. It transcribes text and can reformat it into a checklist, email, or SOP draft.

Charts, diagrams, and custom visuals in chat

Since early 2026, Claude can build diagrams, charts, and interactive visuals directly in the conversation using HTML and SVG—beta on Claude web and desktop. You can iterate in plain language instead of exporting static AI art. This is code-based output, not pixel generation from a diffusion model.

SVG and artifact-style graphics

Ask for flowcharts, simple icons, or labeled diagrams as SVG. Claude writes vector code you can paste into docs or slide decks. Quality depends on prompt clarity; complex brand illustrations still belong in Figma or a generator.

Claude Design (layouts and slides, not AI photos)

Anthropic Labs shipped Claude Design in April 2026 as a research preview for Pro, Max, Team, and Enterprise subscribers. It helps you produce slides, one-pagers, and prototypes from prompts or uploads, with optional design-system import. It composes visual work—it does not replace Midjourney for photorealistic images.

Image generation prompts and MCP connectors

Claude writes strong prompts for Midjourney, Firefly, and similar tools when you ask for tool-specific syntax. To generate pixels inside Claude, connect an image model through an MCP server (remote connectors are supported on paid tiers; the free tier allows limited remote MCP). Claude calls the tool; the external model renders the image.

Prompt engineering for your content stack

Teams publishing GEO-focused guides often pair Claude’s analysis with separate generators, then track which pages earn mentions in AI answers—patterns we summarize in GEO case studies for 2026.

Quick comparison: Claude vs image generators

Frequently Asked Questions

Can Claude make images?

No. Claude does not generate photos or illustrations on its own. Use Midjourney, ChatGPT with GPT Image, Adobe Firefly, Gemini Imagen, or Stable Diffusion—or connect one of those through MCP.

Can Claude read or analyze images I upload?

Yes. Claude analyzes photos, screenshots, documents, and charts on Free, Pro, Max, Team, and Enterprise plans, subject to your plan’s usage limits.

Can Claude create diagrams and charts in chat?

Yes. Claude can build diagrams, charts, and interactive visuals with HTML and SVG in conversation. Anthropic lists this as beta on Claude web and desktop as of March 2026.

What is Claude Design?

Claude Design is an Anthropic Labs product for slides, layouts, and prototypes from prompts or uploads. It is a research preview for paid Claude plans—not a replacement for a diffusion image generator.

Can Claude generate images if I add an MCP connector?

Yes. When you attach an MCP server that exposes image generation, Claude can call that tool and return rendered images in the thread. Rendering still happens on the connected service, not inside Claude’s core model.

Should I use Claude or ChatGPT for visuals?

Use ChatGPT when you want native image generation in one chat. Use Claude when you need stronger document reasoning, code-based diagrams, or a workflow that combines vision analysis with external tools via MCP.

Claude rate limits are the single most complained-about aspect of the product. A viral Reddit post on the topic received over 1,060 upvotes. This guide explains what the limits are at every plan tier, why they exist, and every community-tested strategy for getting more out of your plan before hitting the wall.

1. Anthropic API Usage Tier Qualifications & Thresholds

Your API account’s rate limits are spend-based. The current system has five tiers — Evaluation, Start, Build, Scale, and Custom — and your tier follows your monthly spend level rather than a deposit balance:

Per-model RPM and TPM limits for each tier are published in the Console and in Anthropic’s rate-limits documentation. This table replaces the retired deposit-based Tier 1–4 table ($5 deposit, cumulative-spend thresholds) in November 2026.

2. Requests Per Minute (RPM) and Tokens Per Minute (TPM) by Model

Rate limits apply independently across model families. High-intelligence models (Opus) have tighter token concurrency caps than lightweight models (Haiku):

Anthropic publishes standard per-minute limits for the current usage tiers, read from Anthropic’s rate-limits documentation in October 2026. Limits are requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM), enforced per model class at the organization level:

Model	Start	Build	Scale
Claude Opus 5.5	1,000 / 2,000,000 / 400,000	5,000 / 5,000,000 / 1,000,000	10,000 / 10,000,000 / 2,000,000
Claude Sonnet 5.5	1,000 / 2,000,000 / 400,000	5,000 / 5,000,000 / 1,000,000	10,000 / 10,000,000 / 2,000,000
Claude Haiku 5.5	1,000 / 2,000,000 / 400,000	5,000 / 5,000,000 / 1,000,000	10,000 / 10,000,000 / 2,000,000
Claude Fable 5.1 and Fable 5 (combined)	1,000 / 500,000 / 100,000	2,000 / 1,500,000 / 300,000	4,000 / 4,000,000 / 800,000

Models still available from the prior generation (Sonnet 5, Opus 5, Haiku 4.5) carry the same standard Start limits. New organizations may begin in an Evaluation tier with lower limits that Anthropic does not publish in the standard table, and the Custom tier’s limits are arranged with Anthropic. Your Console Rate limits page shows your account’s actual tier and limits.

3. Diagnosing and Handling HTTP 429 Rate Limit Errors

When your application exceeds either its Requests-Per-Minute or Tokens-Per-Minute cap, the Anthropic API responds with an HTTP 429 Too Many Requests error containing response headers detailing when capacity will reset:

• retry-after: Number of seconds to wait before retrying.
• anthropic-ratelimit-requests-remaining: Remaining requests available in the current 60-second window.
• anthropic-ratelimit-tokens-remaining: Remaining token budget available in the current window.
• anthropic-ratelimit-tokens-reset: ISO timestamp indicating when the token pool will fully refresh.

Production Rate Limit Mitigation Playbook

1. Exponential Backoff with Full Jitter: Never retry immediately in a tight loop. Implement an exponential backoff formula with randomized jitter to prevent thundering herd spikes on your backend.
2. Utilize Prompt Caching: Cached prefix tokens read from memory bypass standard token generation latency and dramatically streamline token processing windows. Read our full Claude AI Pricing and Token Rates Guide for complete caching cost structures.
3. Route Heavy Jobs to the Batch API: For bulk processing, offline report generation, and data extraction, use the Anthropic Messages Batch endpoint. Batch jobs run against separate capacity pools, avoiding live interactive rate caps while cutting token costs by 50%.

---

Why Rate Limits Exist

Claude’s rate limits are primarily about compute capacity, not money. Running Claude Opus 4.8 on complex tasks requires enormous GPU resources. Anthropic limits usage to ensure consistent performance for all users. The limits are enforced per rolling time window, not per calendar day.

Rate Limits by Plan

Free Plan

Access to Claude Sonnet 5 with limited daily usage. Heavy users hit limits after 5-10 substantive prompts. Anthropic adjusts dynamically based on system load.

Claude Pro ($20/month)

Roughly 5x the usage of free. Community consensus: approximately 12 heavy prompts per session before throttling. Light prompts run much longer before hitting limits.

Claude Max 5x ($100/month)

Approximately 5x Pro limit. Anthropic publishes no per-window token figures; paid plans carry a five-hour session limit and a weekly usage limit across all models.

Claude Ultra ($200/month)

10x the Pro weekly limit, with no 5-hour limit for verified accounts. The top tier for developers running Claude Code for extended sessions and professionals processing large document volumes daily.

API Rate Limits (Current Usage Tiers)

API limits are measured in requests per minute (RPM), input tokens per minute (ITPM), and output tokens per minute (OTPM), enforced per model class at the organization level. The current usage tiers and their monthly spend caps are in section 1 above; the standard per-minute limits are in section 2. Your tier advances automatically with usage history and account standing — not by deposit or cumulative credit purchases; the deposit-based Tier 1–4 system is retired. Earlier versions of this page reproduced the retired system’s tables in this section; they were removed in the November 5, 2026 rebuild because the system they describe no longer exists, and the retired figures remain available in this page’s revision history.

Cache-aware ITPM: for current models, only uncached input tokens count toward your ITPM limit — cache_read_input_tokens do not. With an 80% cache-hit rate against a 2,000,000 ITPM limit you can effectively process ~10,000,000 total input tokens per minute, so prompt caching is the single best lever for raising effective throughput.

When you hit a limit, the API returns a 429 with a retry-after header (seconds to wait), plus anthropic-ratelimit-* headers showing remaining requests/tokens and reset times. Limits use a token-bucket algorithm — capacity replenishes continuously rather than resetting at a fixed clock time. The Message Batches API and Managed Agents endpoints have their own separate limits.

When are Claude’s usage limits on the Pro plan?

Claude Pro usage limits operate under a two-tiered allowance structure: a short-term rolling session window that resets every 5 hours, and an overarching weekly cap that resets once per week at a fixed schedule specific to your account:

• 5-Hour Rolling Session Window: Pro plan users receive at least 5 times the usage capacity of the free tier per session. Your 5-hour limit timer begins when you send your first message in an active session block, and full session capacity replenishes continuously 5 hours later.
• Weekly Account Cap: In addition to the 5-hour session allowance, Pro accounts are subject to a weekly usage limit that governs total consumption across all models, Projects, and tools (including Claude Code and Cowork). This limit resets at a fixed, recurring day and time assigned to your account, which you can monitor under Settings > Usage.
• Factors That Consume Limits Fastest: Anthropic’s documentation highlights three primary drivers of rapid usage consumption:
  1. Model choice: Flagship reasoning models like Claude Opus consume your session budget substantially faster than everyday models like Sonnet or Haiku.
  2. Conversation length: Because Claude re-processes the entire conversation history with every new message, long extended threads consume progressively larger token volumes on each subsequent turn.
  3. File attachments: Attaching large codebases, PDFs, or data files significantly increases the prompt token size sent on every turn within that chat.

Where Anthropic does not publish fixed message counts, capacity scales dynamically based on real-time platform traffic. When limits are reached, users can either wait for the rolling reset or upgrade to Claude Max for 5x expanded session and weekly headroom.

Community-Tested Workarounds

• Use Projects with persistent system prompts — reduces token overhead per conversation
• Use Sonnet for routine tasks, Opus 4.8 for complex ones, and Fable 5 for the most demanding work — don’t burn your limit budget on tasks Sonnet handles equally well
• Batch related work into single long sessions — starting five conversations uses more overhead than one long one
• Compress your inputs — extract only relevant sections from long documents before pasting
• Use the API for high-volume predictable workflows — more limit-efficient than the consumer interface for automated tasks

Frequently Asked Questions

How many messages can I send on Claude Pro?

No published exact number — depends on message complexity. Community estimates suggest roughly 12 heavy messages per session before throttling begins on Pro.

Do Claude rate limits reset daily?

Rate limits use a rolling time window, not a fixed midnight reset.

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