Claude Opus 5.5 is Anthropic’s model for long-running coding and knowledge work, released 22 September 2026 as the first model in the Claude 5.5 family. Anthropic’s pitch is efficiency: performance near Claude Fable 5.1 on most work, at a lower price than Opus 5, with output it says is more than 30% faster.

API pricing is $4 per million input tokens and $20 per million output tokens. The model ID is claude-opus-5-5. Context is 1 million tokens, max output 128,000, and the knowledge cutoff Anthropic lists is June 2026. It is the model Claude Duet runs on in the research preview — you cannot swap it for Fable or Sonnet inside Duet.

Numbers below are from Anthropic’s Introducing Claude Opus 5.5 post and the Opus 5.5 model page. Benchmark margins at this level are noisy; treat them as Anthropic’s reported scores, not a lab we re-ran.

Quick Answer: What Is Claude Opus 5.5?

Opus 5.5 is Anthropic’s current default recommendation for most workloads: agentic coding and knowledge work, cheaper and faster to serve than Opus 5.

Table of Contents

• Quick Answer: What Is Claude Opus 5.5?
• What Changed from Opus 5
• Pricing
• How Anthropic Says It Performs
• Opus 5.5 vs Fable 5.1 vs Sonnet 5.5
• Opus 5.5 and Claude Duet
• What to Watch If You Migrate
• Frequently Asked Questions
• Conclusion
• Sources

What Changed from Opus 5

Anthropic describes Opus 5.5 as a step up from Opus 5 on complex work, not a new brand with a mascot. Three changes matter if you already pay for Claude:

Cost and speed. At default settings, Anthropic’s tests put typical workloads about 40% cheaper than Opus 5. List prices are 20% lower ($4/$20 versus $5/$25). Cache reads — the bulk of agent and coding cost — drop from $0.50 to $0.20 per million tokens, which Anthropic calls a 60% cut. Output is more than 30% faster than Opus 5.

Usage. Along with the price cut, Anthropic increased five-hour limits on Pro, Max, Team, and seat-based Enterprise, and added a rate-limit reset subscribers can save and use later. That reset is separate from the one-time Duet preview reset described in Claude Duet explained.

Writing. Early testers told Anthropic the prose is clearer and leads with the important point. That is a product claim, not a writing contest we scored. If you draft for readers rather than for an agent loop, still edit in Write with AI.

Anthropic also says Opus 5.5 posted its strongest automated behavioral-audit scores to date, is less likely than recent models to take hard-to-reverse actions, and is more resistant to prompt injection than Opus 5. Safeguards are in the same family as Fable 5.1 for biology and cybersecurity. Vetted labs apply through Anthropic’s verification programs; this post is not a guide to those programs.

Pricing

Prices are per million tokens, from Anthropic’s launch table and the current pricing docs.

Fast mode in Claude Code and the Claude Platform is up to 2.5× speed at $8 / $40 per million input / output tokens. The Batch API is 50% off input and output.

Subscription chat (Pro, Max) is not billed per token in the same table. If you only use claude.ai, the practical change is higher five-hour limits and whichever model the product selects — including Duet, which locks Opus 5.5.

How Anthropic Says It Performs

Anthropic leads with agentic coding, computer use, and knowledge work, then warns that benchmark gaps are a weaker guide to day-to-day differences than they used to be. Its own use found the gap versus Fable 5.1 narrower than the table suggests. Scores below are Anthropic’s, mostly at max effort unless noted. When safeguards intervened, some cybersecurity, biology, and frontier-model tasks were completed by older Opus models, which Anthropic says likely lowers Opus 5.5’s score on those sets.

A few cost claims Anthropic pairs with those scores:

• At default (medium) effort on FrontierCode, Opus 5.5 beats the GPT-6 Astra figure Anthropic cites at roughly 20% of the cost per task.
• On Terminal-Bench 4.0 it matches Astra’s reported score at about 40% of the cost.
• On Zapier’s AutomationBench, Astra still scores higher (41.4% vs 40.0%). Zapier ran Opus 5.5 without fallback models, so safeguard blocks counted as failures.

For Astra itself, see GPT-6 Astra explained. Do not treat a single coding bench as “Opus replaced Astra.”

Coding anecdotes in the launch post — a large migration finished in a day, an HAProxy C-to-Rust port in 9.5 hours versus 12 for Fable — are Anthropic and early-tester reports. They show the kind of job Opus 5.5 is aimed at: long, multi-repo work where token count is the bill.

Opus 5.5 vs Fable 5.1 vs Sonnet 5.5

Anthropic’s models overview is the decision rule:

• Opus 5.5 — start here for most workloads.
• Fable 5.1 — demanding reasoning and long-horizon agents, or when your own evals still fail on Opus 5.5 at higher effort. Slower, and $10/$50.
• Sonnet 5.5 — speed plus intelligence at $2/$10. The Opus 5.5 announcement said Sonnet 5.5 and Haiku 5.5 would follow; Anthropic’s current lineup lists both as available.
• Haiku 5.5 — high-volume classification, extraction, and routing. From $0.10 / $0.50, with a higher rate above 100,000 prompt tokens.

All four take text and images, and support tools. Opus 5.5 thinking cannot be switched off. Default effort is medium.

Opus 5.5 and Claude Duet

Claude Duet does not offer a model picker. Anthropic’s help center says Duet runs on Opus 5.5 and decides how much effort a job needs. The launch post’s phrase is “designed around” Opus 5.5. Either way, Pro and Max subscribers on the US waitlist are not choosing Fable inside that app.

That matters for the bill. Duet’s morning briefs and background checks draw the same plan limits as the rest of Claude. A cheaper, faster model is part of why Anthropic can leave an agent checking accounts every few hours without a separate Duet price — until those checks burn the five-hour or weekly cap. Larger Duet jobs (a financial model, a deck) still take many turns.

If you want computer-use comparisons on the OpenAI side, read ChatGPT Dots explained. Dots are tied to Astra, not to Opus.

What to Watch If You Migrate

Anthropic lists breaking changes for code that already calls Opus 5. Read the migration guide before you swap the model string. The ones called out on the model page:

• Thinking cannot be disabled.
• Forced tool use returns an error.
• Thinking blocks are tied to the model and the conversation.
• On the Claude API and Google Cloud, the older computer_20251124 computer-use tool is not accepted.
• Text between tool calls can come back inside thinking blocks that look empty at the default display setting, so a UI that streamed that text as progress goes quiet until you set display.

Fast mode is optional and doubles the token price. It is not the default Duet experience.

Frequently Asked Questions

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic’s September 2026 model for long-running coding and knowledge work. It is priced at $4 per million input tokens and $20 per million output tokens, with a 1 million token context window.

How much does Opus 5.5 cost?

Base API price is $4 / $20 per million input / output tokens. Cache reads are $0.20 per million. Fast mode is $8 / $40. Batch requests are 50% off input and output. Chat subscriptions are plan limits, not this per-token table.

Is Opus 5.5 better than Fable 5.1?

Anthropic says Opus 5.5 matches Fable 5.1 on most work and costs less, and that the real-world gap is narrower than some benchmarks. Use Fable 5.1 when a task still fails on Opus 5.5 at higher effort. Fable is $10 / $50 per million tokens.

Is Opus 5.5 the same as Claude 5.5?

Opus 5.5 is the first model in the Claude 5.5 family. Sonnet 5.5 and Haiku 5.5 are the faster, cheaper siblings on Anthropic’s current model list.

Does Claude Duet use Opus 5.5?

Yes. In the research preview, Claude Duet runs on Opus 5.5 and does not let you pick another model. See Claude Duet explained.

When was the knowledge cutoff?

Anthropic lists June 2026 as the reliable knowledge cutoff for Opus 5.5.

Can I use Opus 5.5 on AWS or Azure?

Yes. Anthropic says Opus 5.5 is on Amazon Web Services, Google Cloud, and Microsoft Azure, as well as the Claude Platform (claude-opus-5-5).

Conclusion

Claude Opus 5.5 is Anthropic’s “start here” model: close to Fable on everyday agent and knowledge work, cheaper and faster than Opus 5, and the fixed brain inside Claude Duet. The headline is not a single benchmark. It is fewer tokens per task at a lower price, with thinking left on.

Check the Opus 5.5 announcement before you quote a score in a deck — those tables move. For the agent product that cannot switch off this model, read Claude Duet explained. For the OpenAI flagship people compare it with, read GPT-6 Astra explained.

Sources

• Anthropic — Introducing Claude Opus 5.5
• Anthropic — Claude Opus 5.5 model overview
• Anthropic — Models overview
• Anthropic — Pricing