Claude Sonnet 5.5: How to Use It, Limits & Best Tasks

Everyday AI Guides · AI Tools & Beginner Guides

Reviewed: November 3, 2026. Specs, prices and effort guidance checked against Anthropic’s Claude Platform documentation and the Sonnet 5.5 announcement on this date. Plans, limits and settings can change.

Quick Answer: What Is Claude Sonnet 5.5?

Claude Sonnet 5.5 (searchers often type claude sonnet 5 5) is Anthropic’s mid-tier model, released September 28, 2026. It has a 1 million-token context window, up to 128,000 tokens of output, adaptive thinking and a June 2026 knowledge cutoff. API pricing is $2 per million input tokens and $10 per million output tokens, half the price of Opus 5.5. It is included on the free Claude plan.

Best default: leave effort at Medium in the Claude apps (that is the app default) and raise it to High only when an answer misses real reasoning, not missing context. In the API, Sonnet 5.5 defaults to High.

Claude Sonnet 5.5 Specs at a Glance

Spec	Claude Sonnet 5.5	What it means for you
Release date	September 28, 2026	Newest Sonnet; Anthropic says it won’t be retired before September 28, 2027.
API model ID	`claude-sonnet-5-5` (Amazon Bedrock: `anthropic.claude-sonnet-5-5`)	Swap the ID, then review the breaking changes below before shipping.
Context window	1M tokens (roughly 555,000 English words)	Fits a large codebase or document set, but you still need to say which sources matter.
Max output	128K tokens; up to 300K on the Message Batches API with the `output-300k-2026-03-24` beta header	Long reports and long code changes are possible in one response.
Inputs / outputs	Text and images in, text out	No image generation or audio.
Thinking	Adaptive (the model decides how much to think, guided by effort)	Effort, not a token budget, is your main control.
Knowledge cutoff	June 2026	Use web search or your own files for anything newer.
Default effort	Medium in Claude apps and Claude Code; High on the Claude API	The same prompt can behave differently in the app and the API.

Anthropic reports that Sonnet 5.5 writes output more than 30% faster than Sonnet 5 and costs up to 30% less per completed task in its testing, at the same per-token price. Those are Anthropic’s measurements, not Designs24hr benchmarks. The savings come from using fewer tokens and attempts, so your real cost still depends on your workload and effort setting.

How to Use Claude Sonnet 5.5

Pick the model. In Claude on the web, desktop, iOS or Android, open the model picker and choose Sonnet 5.5. Free accounts get Haiku and Sonnet; Pro, Max and Ultra plans also unlock Opus and Fable.
Set effort in the same picker. The app default is Medium. Leave it there for writing, summaries and normal coding questions.
Ask for a deliverable, not a topic. “Turn these notes into a one-page decision memo with three options and a recommendation” works better than “tell me about these notes.”
Attach the real source material. The 1M-token window only helps if the documents, data or code are actually in the chat. Label which file is authoritative.
Say what must not change and what to flag. For example: “Keep all numbers exactly as written. Mark anything you are unsure about as UNVERIFIED.”
Check the result before you use it. Spot-check figures, quotes and code behavior against the source.

Pattern that works: outcome → context → constraints → format → how you’ll check it. If you can’t say how you’ll check the answer, the task is not scoped tightly enough yet.

Which Sonnet 5.5 Effort Level Should You Choose?

Effort tells Claude how thoroughly to work: how much it thinks, how many tool calls it makes and how long its answer is. Sonnet 5.5 supports five levels. Anthropic recalibrated them for this model, so a level you liked on Sonnet 5 may not behave the same way here.

Effort	Use it for	Anthropic’s guidance
Low	Quick rewrites, extraction, classification, high-volume routine jobs	Good for latency-sensitive chat.
Medium	Everyday writing and analysis; well-specified agentic coding and tool use	Recommended start for well-specified agentic work; app default.
High	Hard reasoning, tricky debugging, multi-source analysis	API default; recommended for general API use and harder agentic coding.
Xhigh	Long-running coding or agent runs (30+ minutes)	Use only if your own tests show a quality gain.
Max	The hardest problems where cost and latency don’t matter	Same: only when evaluations justify it.

Before you raise effort, ask one question: was the answer weak because Claude lacked reasoning, or because the brief was vague or a file was missing? More effort can’t recover information you never provided. Fix the prompt first, then go up one level.

Claude Sonnet 5.5 API Pricing

Usage (per million tokens)	Sonnet 5.5	Opus 5.5
Input	$2	$4
Output	$10	$20
Prompt cache write (5-minute)	$2.50	$5
Prompt cache write (1-hour)	$4	—
Prompt cache read	$0.10	$0.20
Message Batches API	50% off standard input and output rates

Worked example: a job that sends 200,000 input tokens and gets 20,000 output tokens back costs about $0.40 + $0.20 = $0.60 on Sonnet 5.5, or $1.20 on Opus 5.5. If the same 200,000-token document is reused from cache, the input portion of later calls drops to about $0.02. Batch the job instead and the first call falls to about $0.30.

These are API rates. They are separate from Claude Free, Pro, Max or Ultra subscriptions, which use plan usage limits instead of per-token billing.

Claude Sonnet 5.5 vs Opus 5.5: Which Should You Use?

Anthropic positions Sonnet 5.5 for well-scoped everyday work, coding, documents, slides and spreadsheets, and Opus 5.5 for harder, open-ended work that needs sustained judgment. Opus 5.5 costs twice as much per token and defaults to Medium effort in the API.

Task	Start with	Why
Rewrite, summarize or restructure a document	Sonnet 5.5	Clear brief, fast iteration, lower cost
Bug fix with a clear repro, or a small feature	Sonnet 5.5	Easy to review the diff
High-volume support replies or extraction	Sonnet 5.5 at Low/Medium	Speed and token cost dominate
Large multi-file refactor in an unfamiliar repo	Opus 5.5 (or escalate)	Longer reasoning chain, costly mistakes
Research with conflicting sources and high stakes	Opus 5.5	Harder judgment across long context

Simple escalation rule: write down what a correct result must contain, run Sonnet 5.5 with a tight brief, and fix the prompt once if something was unclear. If the same requirement is still missed, run Opus 5.5 against the same checklist and keep whichever output passes. Opus won’t always win, and many jobs never need it.

Developer Notes: Moving From Sonnet 5

Sampling parameters are rejected. Setting temperature, top_p or top_k returns a 400 error. Remove them from requests.
Lightest thinking setting. between_tools limits thinking to the gaps between tool calls. It works with low, medium and high effort and returns a 400 at xhigh or max.
Re-tune effort. Levels were recalibrated, so rerun your evals at Medium and High before copying your Sonnet 5 setting.
Check all five breaking changes in Anthropic’s Sonnet 5.5 overview before swapping the model ID in production.
Some security requests fall back. Anthropic says certain higher-risk cybersecurity requests are answered by Sonnet 5 instead, and this is shown in the response.

3 Sonnet 5.5 Prompts That Are Easy to Check

To copy a prompt, highlight the text in the box and press Ctrl+C (Cmd+C on Mac). Replace the bracketed parts.

Notes → Action Plan

Turn the notes below into an action plan for this goal: [GOAL]. 1. List decisions already made. 2. Separate confirmed facts from assumptions. 3. List next actions in priority order with owner and deadline where the notes say so. 4. Flag missing information instead of inventing it. Notes: [PASTE NOTES]

Long Document Question

Using only the attached files as the source of truth, answer: [QUESTION] Give the direct answer first, then the strongest supporting quotes with file names, then any evidence that limits the answer. If the files don’t answer it, say so.

Scoped Bug Fix

Bug: [WHAT HAPPENS] Expected: [WHAT SHOULD HAPPEN] Repro: [STEPS] Find the root cause, propose the smallest fix, and list the tests that prove it works. Don’t change public function signatures or unrelated files.

Claude Sonnet 5.5 Troubleshooting

Problem	Likely cause	Fix
Can’t find Sonnet 5.5 in the app	App out of date, or a workspace admin limits models	Update the app or refresh the web page; in a managed workspace, ask your admin.
Can’t find the Effort control	Looking in Settings	Effort lives in the model picker next to the model name.
Answers feel slow	Effort set to High or above	Drop to Medium or Low for chat and quick edits.
Answers are shallow on hard problems	Vague brief or Low effort	Add success criteria and sources first; then raise effort one step.
API returns 400 after switching models	Sampling parameters or `between_tools` at xhigh/max	Remove temperature/top_p/top_k; use low–high effort with `between_tools`.
Output stops early	Hit the 128K output cap	Split the job, or use the Batch API’s 300K beta for long offline jobs.
Hitting usage limits fast	Plan limits plus high effort and huge attachments	Use lower effort, attach only needed files, start new chats for new tasks.

Model limits and plan limits are different things. The 1M context and 128K output are model specs. How many messages you can send depends on your Claude plan and current usage rules.

Related Claude Guides

Claude in Chrome: setup and permission controls for browser tasks.
Claude Slides for turning Sonnet output into a presentation.
Transfer ChatGPT memory to Claude if you’re switching assistants.
All AI Tools & Beginner Guides

Claude Sonnet 5.5 FAQ

Is Claude Sonnet 5.5 free?

Yes, within limits. The free Claude plan includes Sonnet and Haiku. Paid plans raise usage limits and add Opus and Fable. API use is billed per token.

What is the Sonnet 5.5 context window?

1 million tokens of input context, with up to 128,000 output tokens per response (300,000 on the Batch API with the beta header).

Is Sonnet 5.5 better than Opus 5.5?

Not overall. Sonnet 5.5 is faster and half the price per token. Opus 5.5 is the stronger choice for the hardest open-ended, long-horizon work.

What effort level should I use?

Medium for most chat and well-defined tasks, High for difficult reasoning or debugging. Use Xhigh or Max only when testing shows they help.

Does Sonnet 5.5 know about recent events?

Its knowledge cutoff is June 2026. For anything newer, turn on web search or give it current sources.

Sources Checked

Anthropic Claude Platform — Claude Sonnet 5.5 overview: model ID, context, output limits, pricing, cutoff, breaking changes.
Anthropic Claude Platform — Effort: levels, defaults and workload guidance.
Anthropic — Introducing Claude Sonnet 5.5: positioning and Anthropic-reported speed and cost changes.

Editorial note: Designs24hr is not affiliated with or endorsed by Anthropic. Product specifications and performance claims attributed to Anthropic come from its current documentation and release materials. This guide does not claim independent benchmark testing. Verify important outputs before using them for consequential decisions.