# Claude Opus 5.5 vs Sonnet 5.5: which to use for coding Authors: AutoRouter Author profile: https://autorouter.com/about Published: 2026-09-22T17:10:00.000Z Updated: 2026-11-10T21:00:00.000Z Reading time: 12 min Canonical: https://autorouter.com/blog/claude-opus-5-5-vs-sonnet-5-5 Publisher: AutoRouter, the Mac app that routes Codex, Claude Code and other AI tools to every model (https://autorouter.com) About AutoRouter: a Mac app (macOS 14 or later) that adds every leading model to Codex, Claude Code, Claude Desktop, T3 and Terminal and picks the model for each task, using your ChatGPT and Claude plans first. Install from https://autorouter.com/AutoRouter.dmg, or ask your agent to follow https://autorouter.com/setup.md. Routing across your plans is free; Pro is $50 a month with $20 of included usage. Unrelated to OpenRouter's Auto Router. > Start coding work on Claude Opus 5.5, released September 22, 2026 and Claude Code's default model, interface design and animation included, and move a task down to Claude Sonnet 5.5 when it is well scoped, such as a bug fix or a feature built against a spec. As of November 10, 2026, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, twice Sonnet 5.5's $2 and $10. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.074 on Opus 5.5 and $0.037 on Sonnet 5.5. ## Key facts - As of November 10, 2026, Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, twice Claude Sonnet 5.5's $2 and $10, and cache reads cost $0.20 and $0.10 per million tokens ([Anthropic](https://platform.claude.com/docs/en/about-claude/pricing)). - One assumed agent request (45,000 cached input tokens, 5,000 new input tokens and 2,000 output tokens) costs $0.074 on Claude Opus 5.5 and $0.037 on Claude Sonnet 5.5 at list prices on November 10, 2026, by AutoRouter's arithmetic. - Anthropic's November 9, 2026 guide starts most coding work on Claude Opus 5.5 at medium effort and moves well-scoped work, such as a bug fix or a feature built against a spec, down to Claude Sonnet 5.5 ([Anthropic](https://claude.dev/blog/which-claude-model-should-you-build-with/)). - On Cursor's CursorBench 4.0, as of November 10, 2026, Claude Opus 5.5 at medium effort scored 52.5% at $2.91 per task and Claude Sonnet 5.5 at medium effort scored 39.2% at $0.52 per task ([Cursor](https://cursor.com/cursorbench)). - In AutoRouter's measurement of five small Claude Code tasks on October 28, 2026, an average run cost $0.110 on Claude Opus 5.5 and $0.049 on Claude Sonnet 5.5 as Claude Code priced them on a Max plan, and all 15 runs on each model passed their checks ([data](/blog/claude-code-task-costs-measured)). As of November 10, 2026, start coding work on Claude Opus 5.5, and move a task down to Claude Sonnet 5.5 when it is well scoped: a bug fix, or a feature built against a spec. That is [Anthropic's own advice](https://claude.dev/blog/which-claude-model-should-you-build-with/), and Opus 5.5 is Claude Code's [default model](https://code.claude.com/docs/en/model-config#default-model-setting) on every plan. It is also the Claude model for visual work: building an interface from a design, reading charts and screenshots, and animation. Opus 5.5 costs [$4 per million input tokens and $20 per million output tokens](https://platform.claude.com/docs/en/about-claude/pricing), twice Sonnet 5.5's $2 and $10. [AutoRouter](https://autorouter.com), the Mac app for Codex and Claude Code, wrote this post and worked the prices through: one assumed agent request costs $0.074 on Opus 5.5 and $0.037 on Sonnet 5.5. Every outside figure below comes from the maker's or publisher's own page, linked where it is used and checked on November 10, 2026. ![AutoRouter's launch poster for Claude Opus 5.5. It reads: On Tuesday, Sep 22, Anthropic released Claude Opus 5.5. A workhorse that's also good at visual design and animation. Now in AutoRouter. Opus 5.5. The picture is a tower of stacked colonnades that grows as the view circles it, drawn in white tokenizer tokens on a green ground.](https://autorouter.com/models/claude-opus-5-5.png) ## What is Claude Opus 5.5 good at? Claude Opus 5.5 is the model Anthropic builds for long-running agentic coding and knowledge work, [released on September 22, 2026](https://www.anthropic.com/claude-opus-5-5) as the first of the Claude 5.5 family. Its [model page](https://platform.claude.com/docs/en/models/opus-5-5/overview) lists multihour coding agents, large-scale refactoring and complex systems engineering among its typical uses. Against Opus 5, its predecessor, Anthropic's announcement says input and output tokens cost 20% less, cache reads cost 60% less, typical workloads cost 40% less at default settings in its own tests, and output arrives more than 30% faster. [Claude Sonnet 5.5 followed on September 28, 2026](https://www.anthropic.com/claude-sonnet-5-5), positioned as the quicker, cheaper companion for well-scoped everyday tasks and bug fixes. Opus 5.5 does nearly every kind of coding work well, and its visual work stands out. As of November 10, 2026, this is what Anthropic's pages say about it, with one example of our own: - **Interfaces and visual polish.** Anthropic's [launch post](https://www.anthropic.com/claude-opus-5-5) reports that when a tester had several Claude models build a game from one prompt, Opus 5.5 scored highest for its graphics and polish. Its [prompting guide](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5#frontend-design-defaults) adds a caution: asked for frontend work with no design direction, Opus 5.5 falls back on a few default styles, and it responds well when you name the specific patterns to avoid. - **Charts, diagrams and screenshots.** The same [guide](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5) says Opus 5.5 reads visual material more accurately than Opus 5 without extra tooling, including which boxes an arrow connects in a flowchart and what changed between two versions of a diagram. Anthropic's [guide to choosing a model](https://claude.dev/blog/which-claude-model-should-you-build-with/) starts computer use and vision-heavy work on Opus 5.5. - **Animation.** Anthropic publishes no measurement of animation work. Our own example is the loop at the top of this post: Claude Opus 5.5 built and animated it in code in Claude Code, stacking a 3D scan of a Roman colonnade into a spiral tower that never stops growing. ## Which coding work belongs on Opus 5.5, and which on Sonnet 5.5? Complex, open-ended and long-running work belongs on Claude Opus 5.5, and well-scoped work belongs on Claude Sonnet 5.5, most of all when the task comes with a spec and a test that checks the result. That split is Anthropic's. Its [guide to choosing a model](https://claude.dev/blog/which-claude-model-should-you-build-with/) (November 9, 2026) recommends starting most coding work on Opus 5.5 at `medium` effort, its [guide to building with Sonnet 5.5](https://claude.dev/blog/building-with-claude-sonnet-5-5/) puts scoped, checkable tasks on Sonnet 5.5, and its [Sonnet 5.5 announcement](https://www.anthropic.com/claude-sonnet-5-5) says Opus 5.5 "remains clearly stronger at complex, open-ended work requiring sustained judgment". | Kind of work | Start with | Why (Anthropic's guide, November 9, 2026) | | ----------------------------------------------------------------------------- | ------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------- | | Everyday engineering and debugging, feature work across several pull requests | Opus 5.5 at `medium`, then `high` | The guide starts this work on Opus 5.5 for its balance of cost and capability. | | A migration or a performance optimization | Opus 5.5 (Fable 5.2 for a major one) | The guide says these can have a clear spec and still be too hard for Sonnet 5.5. | | Computer use and work heavy on screenshots | Opus 5.5 at `medium` | The guide describes these long agentic loops as work Opus 5.5 is built for. | | An interface built from a design or screenshot, visual polish, animation | Opus 5.5 at `medium`, then `high` | The guide starts vision-heavy work on Opus 5.5, and Anthropic's launch post reports its lead on graphics and polish. | | A bug fix, feature iteration, work against a spec | Sonnet 5.5 at `medium`, then `high` | The guide says Sonnet 5.5 handles scoped tasks well at half Opus 5.5's token price. | | Agent tasks you run repeatedly: investigation, review, drafting | Sonnet 5.5 at `medium` | The guide says well-defined, repeated tasks rarely need a bigger model. | | You aren't sure yet | Opus 5.5 at `medium` | It is Claude Code's default model, and Anthropic's [models overview](https://platform.claude.com/docs/en/models/overview) says to start there. | Claude Code's [cost guide](https://code.claude.com/docs/en/costs) puts the starting point differently: it says Sonnet handles most coding tasks well and reserves Opus for complex architectural decisions and multi-step reasoning. Both pages give complex, multi-step work to Opus. [Which model to use for which coding task](/blog/which-model-for-which-task) covers Haiku 5.5, Fable 5.2 and OpenAI's lineup. ## How much does Opus 5.5 cost compared with Sonnet 5.5? Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens as of November 10, 2026, twice Claude Sonnet 5.5's $2 and $10, and the same ratio holds on every line of Anthropic's token price list ([pricing](https://platform.claude.com/docs/en/about-claude/pricing)). | List price, November 10, 2026 | Claude Opus 5.5 | Claude Sonnet 5.5 | | -------------------------------------------- | --------------- | ----------------- | | Input, per million tokens | $4 | $2 | | 5-minute cache write, per million tokens | $5 | $2.50 | | 1-hour cache write, per million tokens | $8 | $4 | | Cache read, per million tokens | $0.20 | $0.10 | | Output, per million tokens | $20 | $10 | | Fast mode, input / output per million tokens | $8 / $40 | Not offered | | Default effort on the API | `medium` | `high` | | Default effort in Claude Code | `medium` | `medium` | The example below prices the assumed request AutoRouter's posts use: 45,000 cached input tokens, 5,000 new input tokens written to the 5-minute cache Claude Code [uses with an API key](https://code.claude.com/docs/en/prompt-caching#which-ttl-each-request-gets), and 2,000 output tokens, thinking included. ```text One assumed agent request, list prices as of November 10, 2026 Claude Opus 5.5 45,000 cached input x $0.20 per million = $0.0090 5,000 new input x $5.00 per million = $0.0250 (5-minute cache write) 2,000 output x $20.00 per million = $0.0400 Total $0.0740 Claude Sonnet 5.5 45,000 cached input x $0.10 per million = $0.0045 5,000 new input x $2.50 per million = $0.0125 (5-minute cache write) 2,000 output x $10.00 per million = $0.0200 Total $0.0370 ``` At this request size, $20 of API usage pays for 270 requests on Opus 5.5 or 540 on Sonnet 5.5. Every price is doubled, so a finished task costs something other than twice as much only when the two models use different numbers of tokens and steps to finish it. ## What does a finished task cost on each model? A finished task cost 2.2 times as much on Claude Opus 5.5 as on Claude Sonnet 5.5 in AutoRouter's October 28, 2026 measurement of small tasks. On Cursor's harder benchmark tasks it cost between 1.9 and 5.6 times as much at matching effort settings, and Opus 5.5 scored higher at every setting. **Small, well-specified tasks.** In [What common Claude Code tasks cost, measured on each model](/blog/claude-code-task-costs-measured), AutoRouter ran five tasks three times each in a 1,176-line Python library on October 28, 2026: a commit message, a rename across files, tests for one module, a failing-test fix and a multi-file refactor. An average run cost $0.110 on Opus 5.5 and $0.049 on Sonnet 5.5 as Claude Code priced them on a Max plan. All 15 runs on each model passed their checks. The study shows what these small tasks cost on each model and that both models finished them. It did not test harder work. **Harder tasks, by effort.** [CursorBench 4.0](https://cursor.com/cursorbench), which Cursor runs on ambiguous, multi-file tasks taken from real Cursor sessions, lists both models at every effort level as of November 10, 2026. Cursor computes cost per task from each model's published token prices. | Effort | Opus 5.5 score | Opus 5.5 cost per task | Sonnet 5.5 score | Sonnet 5.5 cost per task | | -------- | -------------- | ---------------------- | ---------------- | ------------------------ | | `low` | 43.7% | $1.17 | 35.8% | $0.37 | | `medium` | 52.5% | $2.91 | 39.2% | $0.52 | | `high` | 56.0% | $3.97 | 47.8% | $1.20 | | `xhigh` | 56.0% | $6.98 | 53.1% | $2.81 | | `max` | 57.8% | $13.43 | 55.5% | $7.05 | - **At `medium`, where Claude Code starts both models,** Opus 5.5 scored 13.3 points higher and cost 5.6 times as much per task. - **Near $1.20 per task, Sonnet 5.5 scored higher:** 47.8% at `high`, against 43.7% for Opus 5.5 at `low`. - **Near $2.90 per task, the two met:** Sonnet 5.5 at `xhigh` scored 53.1% and Opus 5.5 at `medium` scored 52.5%. - **Above that, Opus 5.5 scored higher for less:** 56.0% at `high` for $3.97, against 55.5% for $7.05 on Sonnet 5.5 at `max`. Cursor notes that small differences in its scores may not be statistically meaningful, and Claude Code's docs say [effort is calibrated per model](https://code.claude.com/docs/en/model-config#adjust-effort-level), so one level name stands for a different amount of work on each. To price your own work, run one hard task from your backlog on both models and compare what `/usage` shows. ## How do Opus 5.5 and Sonnet 5.5 count against a Claude plan? Both models draw on the same five-hour and weekly limits of a Claude plan, and Claude Code [starts every plan on Opus 5.5](https://code.claude.com/docs/en/model-config#default-model-setting) until you change the model. Anthropic says the model you use affects how fast you reach a limit, and doesn't publish by how much, as of November 10, 2026 ([plans](https://claude.com/pricing), [usage limits](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work)). - **Shared limits.** Pro ($20 a month), Max ($100) and Ultra ($200) each have a weekly limit that applies across all models. Pro and Max also have a five-hour limit, which verified Ultra accounts don't. - **What counts toward them.** Anthropic's Help Center lists the length and complexity of the conversation, the features you use, the model and the effort level. It gives no ratio between Opus 5.5 and Sonnet 5.5. - **Limits for one model family.** Claude Code can also stop on an Opus limit or a Sonnet limit. Each applies to that family only, so [switching to the other with `/model`](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit) keeps you working. ## How does AutoRouter choose between Opus 5.5 and Sonnet 5.5? AutoRouter, the Mac app for Codex and Claude Code, [picks a model for each task](https://autorouter.com/#product) while it is the model selected. Your rules go first. Otherwise AutoRouter works out how hard the task is and only considers models that clear the quality bar for it. Hard work goes to the most capable model your plans cover, even when a cheaper one comes close. Everything else goes to the model that costs you least. Unclear work is never treated as a quick task, and a task that continues keeps its model or moves up. In Claude Code, Claude models run on your Claude plan. You can choose Opus 5.5 or Sonnet 5.5 in the picker yourself, which sends every request straight to that model, or write a rule in plain English such as "Use Claude Opus 5.5 for UI work." Every request shows its model, what it cost and why. AutoRouter is in beta and sometimes misjudges how hard a task is; you can rate the pick. It earns 10% on paid models and nothing on your plans. ## Questions ### Is Claude Opus 5.5 better than Sonnet 5.5 at coding? On complex, open-ended work, yes, by Anthropic's own account: its [Sonnet 5.5 announcement](https://www.anthropic.com/claude-sonnet-5-5) says Opus 5.5 is clearly the stronger of the two on work that needs sustained judgment. On [CursorBench 4.0](https://cursor.com/cursorbench), which Cursor runs, Opus 5.5 scored higher than Sonnet 5.5 at every effort level as of November 10, 2026, from 43.7% against 35.8% at `low` to 57.8% against 55.5% at `max`. Sonnet 5.5 is the model Anthropic calls strongest at well-scoped everyday tasks, such as fixing a bug. ### Should you raise Sonnet 5.5's effort or switch to Opus 5.5? Raise effort to `high` first, then switch. Anthropic's [guide](https://claude.dev/blog/which-claude-model-should-you-build-with/) gives the order: check that the model can verify its own work with a test or a build, raise effort one level, then move to the next model up at a lower effort. Its [guide to building with Sonnet 5.5](https://claude.dev/blog/building-with-claude-sonnet-5-5/) says Sonnet 5.5 at `xhigh` or `max` can lose its balance of quality, speed and cost, and to consider Opus 5.5 then. On SWE-bench Pro, the first guide reports Opus 5.5 at `medium` scoring about 2.5 points below `high` for about 70% of the cost. ### Is Claude Opus 5.5 good at visual design and animation? Yes, with direction. Anthropic's [launch post](https://www.anthropic.com/claude-opus-5-5) reports that Opus 5.5 scored highest among Claude models for graphics and polish in a tester's game build, and its [prompting guide](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5#frontend-design-defaults) says to name the specific styles you don't want, because without direction it falls back on a few defaults. Anthropic publishes no measurement of animation work, as of November 10, 2026. The loop at the top of this post is one example: Opus 5.5 built and animated it in code, in Claude Code. ### How do you switch between Opus 5.5 and Sonnet 5.5 in Claude Code? Run `/model sonnet` or `/model opus` in a session. On the Anthropic API the aliases point to Sonnet 5.5 and Opus 5.5, and both start at `medium` effort ([model configuration](https://code.claude.com/docs/en/model-config)). Each model has its own prompt cache, so the first request after a switch [reads the whole conversation with no cache hits](https://code.claude.com/docs/en/prompt-caching#switching-models). On the Claude API, a move up to Opus 5.5 keeps the conversation's reasoning and a move down to Sonnet 5.5 [loses it](https://platform.claude.com/docs/en/models/sonnet-5-5/whats-new-sonnet-5-5). The [`opusplan` alias](https://code.claude.com/docs/en/model-config#opusplan-model-setting) plans with Opus and carries out the plan with Sonnet. Anthropic's [guide](https://claude.dev/blog/which-claude-model-should-you-build-with/) notes that this puts code edits on Sonnet, the reverse of its advice, and says to measure it on your own tasks first. [How to change the model in Claude Code](/blog/claude-code-model-settings) covers every setting. ## Sources - [Introducing Claude Opus 5.5](https://www.anthropic.com/claude-opus-5-5), Anthropic, September 22, 2026 - [Introducing Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5), Anthropic, September 28, 2026 - [Prompting Claude Opus 5.5](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5-5), Claude Platform docs, Anthropic - [Pricing](https://platform.claude.com/docs/en/about-claude/pricing), Claude Platform docs, Anthropic, checked November 10, 2026 - [Which Claude model should you build with?](https://claude.dev/blog/which-claude-model-should-you-build-with/), Anthropic, November 9, 2026 - [Building with Claude Sonnet 5.5](https://claude.dev/blog/building-with-claude-sonnet-5-5/), Anthropic, September 28, 2026 - [Models overview](https://platform.claude.com/docs/en/models/overview), [Claude Opus 5.5](https://platform.claude.com/docs/en/models/opus-5-5/overview) and [What's new in Claude Sonnet 5.5](https://platform.claude.com/docs/en/models/sonnet-5-5/whats-new-sonnet-5-5), Claude Platform docs, Anthropic - [Model configuration](https://code.claude.com/docs/en/model-config), [Manage costs effectively](https://code.claude.com/docs/en/costs), [How Claude Code uses prompt caching](https://code.claude.com/docs/en/prompt-caching) and [Error reference](https://code.claude.com/docs/en/errors#youve-hit-your-session-limit), Claude Code docs, Anthropic - [Plans and pricing](https://claude.com/pricing), Anthropic, checked November 10, 2026 - [How do usage and length limits work?](https://support.claude.com/en/articles/11647753-how-do-usage-and-length-limits-work), Claude Help Center - [CursorBench](https://cursor.com/cursorbench), Cursor, checked November 10, 2026 ## Read next - [Claude Fable 5.2 vs Opus 5.5: which to use for coding](https://autorouter.com/blog/claude-fable-5-2-vs-opus-5-5.md): Keep everyday coding on Claude Opus 5.5 and move to Claude Fable 5.2, released November 9, 2026, for major refactors, migrations, long unattended runs and tasks where Opus 5.5 at high effort still falls short. As of November 10, 2026, Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, twice Opus 5.5's price, with the same $0.20 cache-read price. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.139 on Fable 5.2 and $0.074 on Opus 5.5. - [Gemini 4 Carbon vs Gemini 4 Argon: which to use for coding](https://autorouter.com/blog/gemini-4-carbon-vs-argon.md): Keep everyday coding on Gemini 4 Argon, and bring in Gemini 4 Carbon, generally available since November 10, 2026, for root-cause debugging in large codebases, architecture changes, migrations and security audits. As of November 10, 2026, the two models share one price list: $2 per million input tokens and $10 per million output tokens through November 30, then $4 and $20, and Google says Carbon uses more tokens on each request. By the arithmetic of AutoRouter, the Mac app for Codex and Claude Code, one assumed agent request costs $0.0345 on either model now and $0.069 from December 1. - [Muse Ember vs Claude Haiku 5.5: Meta's low-cost model for coding](https://autorouter.com/blog/muse-ember-vs-claude-haiku-5-5.md): Muse Ember is Meta's smallest Muse model, released on November 9, 2026 at $0.10 per million input tokens and $0.50 per million output tokens, the same list prices as Claude Haiku 5.5 and GPT-6 Luna on prompts up to 100,000 tokens. Meta says it matches Haiku 5.5's best score on the Artificial Analysis Intelligence Index at about a third of the cost per task, and Meta's own docs keep multi-file changes on a larger model. AutoRouter, the Mac app for Codex and Claude Code, has Muse Ember in its model picker. Tags: [Models](https://autorouter.com/blog/tags/models.md), [Claude Code](https://autorouter.com/blog/tags/claude-code.md), [Comparisons](https://autorouter.com/blog/tags/comparisons.md) Drafted with AI from the sources linked above, then checked and edited by the AutoRouter team.