Claude Opus 5.5 and Sonnet 5.5: What Changed and Which Model to Use

Opus 5.5 and Sonnet 5.5 at a glance

Both models read text and images, use tools, and share a June 2026 knowledge cutoff. Anthropic's own advice for developers is to start with Opus 5.5 for most workloads, and to move up to Claude Fable 5.1 only for the most demanding reasoning and long-horizon agent work.

What is new in Claude Opus 5.5

Opus 5.5 is the headline release. Anthropic says it matches Fable 5.1 on most work, costs about 40% less than Opus 5 on typical workloads, and generates output about 30% faster. The list price also fell, from $5 to $4 per million input tokens and from $25 to $20 per million output tokens.

The knowledge-work score is the one most relevant to Cowork. GDPval-style tests grade real deliverables such as reports, analyses, and spreadsheets, and a jump of more than 130 Elo points is large for a single release. Opus 5.5 also scored 81.8% on OSWorld 2.1, a test of operating a computer through its screen, which matters if you use computer use in Cowork.

Safeguards that change behavior

• Cybersecurity requests are routed to Claude Opus 4.8 rather than answered by Opus 5.5, unless the organization is in Anthropic's Cyber Verification Program.
• Biology safeguards match Fable 5.1. Deeper life-sciences research requires the Life Sciences Verification Program.
• Thinking cannot be turned off. Opus 5.5 always uses adaptive thinking; you steer how hard it thinks with the effort setting.
• Fast mode is available on the API at $8 input and $40 output per million tokens for latency-sensitive work.

Opus 5.5 is available on Pro, Max, Team, and Enterprise plans, in Claude Code, in Claude Cowork, and on the Claude API, Amazon Web Services, Google Cloud, and Microsoft Azure.

What is new in Claude Sonnet 5.5

Sonnet 5.5 keeps Sonnet 5's list price of $2 input and $10 output per million tokens but does more for it: Anthropic reports output that is over 30% faster and a cost per task up to 30% lower than Sonnet 5. In Claude's apps and Claude Code, its default effort is Medium; on the API it defaults to High.

Anthropic positions Sonnet 5.5 as strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. That last point makes it a natural fit for Claude Docs and Claude Slides, which launched the same month.

Opus 5.5 vs Sonnet 5.5: benchmarks side by side

The surprise is how close the two are. Sonnet 5.5 edges ahead on Terminal-Bench and is essentially tied with Opus 5.5 on the GDPval knowledge-work test, at half the price. Opus 5.5 keeps a clear lead on FrontierCode, the hardest coding benchmark in the set, which tracks with Anthropic's framing: Opus for complex work requiring careful judgment, Sonnet for well-scoped tasks.

Which model to use for Cowork tasks

In the merged Claude app, the model picker sits where it always has, so you can switch per conversation. A practical default for most people is Sonnet 5.5 for routine work and Opus 5.5 when the task is long, ambiguous, or high stakes.

Both models draw on your plan's usage limits, and a bigger model at higher effort uses more of your allowance per task. If you hit limits often, move routine and scheduled work to Sonnet 5.5 first. Our pricing guide explains how the plans compare, and the merge explainer covers where settings moved in the new app.

Effort matters as much as the model

Both models think adaptively, and the effort setting controls how much. On a well-defined task, raising effort on Sonnet 5.5 is often a cheaper fix than switching to Opus 5.5, and Opus 5.5 at its default medium effort is usually enough for a first draft. If results disappoint, raise effort before switching models.

Switching from Opus 5 or Sonnet 5: a checklist

New models usually arrive as the default for new conversations, but work you set up earlier does not always move with them. Before you assume everything now runs on the new generation, walk through these checks:

1. Open each scheduled task and confirm which model it uses. A task created months ago may still name an older model, and it will keep running on it while that model is available.
2. Rerun your three most common tasks on the new model next to the old one. Compare the result, the time it took, and how much of your usage it consumed. Keep notes; one honest comparison beats a dozen benchmark tables.
3. Reread your instructions and skills. Standing instructions written to work around an older model's habits, such as reminders to double-check a format, may no longer be needed and just add noise.
4. Watch the usage meter for a week. Faster output and lower cost per task should stretch your plan further, but heavier use of the bigger model can cancel that out.
5. Tell your team what changed. If colleagues share skills or projects with you, agree on which model to use for shared recurring work so results stay consistent.

Developers have one extra step: older model IDs keep working until their retirement dates, so nothing breaks the day a new model launches. Pin the new ID in a test environment, run your evaluation set, and only then switch production traffic. Anthropic publishes a migration guide for moving to the new Opus, which lists the parameters that changed between generations.

Where Fable 5.1 and Haiku fit

Haiku 5.5 has been announced for the coming weeks but was not out as of early October 2026. Until it ships, Haiku 4.5 remains the budget option, with a 200K-token context window and a retirement date no sooner than October 15, 2026. Mythos 5.1, Anthropic's most capable model, is limited to vetted organizations; our Fable 5.1 and Mythos 5.1 guide explains the access programs.

Notes for developers and admins

• Model IDs: claude-opus-5-5 and claude-sonnet-5-5 on the Claude API, Microsoft Foundry, and Google Cloud; anthropic.claude-opus-5-5 and anthropic.claude-sonnet-5-5 on Amazon Bedrock.
• Thinking: both use adaptive thinking steered by the effort parameter. The older extended-thinking mode with a fixed token budget is not accepted on these models.
• Caching: Opus 5.5 cache reads cost 5% of the base input price ($0.20); Sonnet 5.5 cache reads cost 10% ($0.20). Batch requests are 50% off.
• Long outputs: both allow 128K output tokens on the Messages API, and up to 300K on the Batches API with a beta header.
• Lifecycle: Anthropic commits to keeping Opus 5.5 until at least September 22, 2027, and Sonnet 5.5 until at least September 28, 2027.

For teams building their own agents on these models rather than using Cowork, our Managed Agents guide covers Anthropic's hosted option. Enterprise admins rolling out new models can start with our enterprise deployment guide.

Frequently asked questions