Claude Opus 5.5 review
Anthropic's top model, cheaper and faster than Opus 5.
Anthropic's most capable model, now cheaper and faster than Opus 5. It leads published tests for coding agents, computer use and office work, and has the highest independent overall score of any model on this site. It is still one of the pricier options per token.
What is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's flagship model, released on 22 September 2026 as the first model of the Claude 5.5 family. It replaces Claude Opus 5 and is built for hard reasoning, coding on large codebases and long tasks that take many steps.
Key features
- Top published score for coding agents on Terminal-Bench 4.0: 66.4%
- Very good at using a computer: 81.8% on OSWorld 2.0
- Reads 1 million tokens at once and writes up to 128,000
- Always thinks before answering, with an effort setting to trade depth for speed and cost
- 20% cheaper per token than Opus 5, with output more than 30% faster
Specs and price
| Made by | Anthropic |
| Released | 22 Sep 2026 |
| API price | $4 per 1M input tokens, $20 per 1M output tokens |
| Batch price | $2 in / $10 out |
| Fast mode | $8 in / $40 out |
| How much it can read | 1,000,000 tokens |
| Longest answer | 128,000 tokens (300,000 in the Batch API) |
| Knows events up to | June 2026 |
| Where | Claude apps, Claude Code, Claude API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry |
| Free to download | No |
| Name in the API | claude-opus-5-5 |
Test scores
| Test | Score | Published by |
|---|---|---|
| Terminal-Bench 4.0Coding agents working in a terminal | 66.4% (GPT-6 Astra: 57.9%) | Anthropic |
| OSWorld 2.0Using a computer like a person | 81.8% | Anthropic |
| GDPval-AA v2.1Real office work (rating) | 1846 (Claude Fable 5.1: 1735) | Anthropic |
| Humanity's Last ExamVery hard expert questions, with tools | 67.7% | Anthropic |
| AutomationBenchAutomating business workflows | 40.0% (GPT-6 Astra leads: 41.4%) | Anthropic |
| Artificial Analysis indexOverall score across many tests | 58 | Independent |
Scores published by Anthropic are the maker's own numbers. We will add our own results in the hands-on review.
How to use Claude Opus 5.5
- Easiest: open claude.ai or the Claude app and choose Opus 5.5 in the model picker. Check Anthropic's plans page for which plans include it.
- In Claude Code, switch to Opus 5.5 with the model picker.
- For developers: call
claude-opus-5-5in the Claude API, or use it through Amazon Bedrock, Google Vertex AI or Microsoft Foundry.
Who it is for
Good fit
- Developers working on large codebases or long agent tasks
- Analysts doing finance or science work with charts and documents
- Teams that want the strongest results and can pay for them
Skip it if
- Simple, high-volume jobs: Claude Haiku 5.5 costs 40 times less per input token
- Apps that turn thinking off or force a specific tool: those settings now return errors
Pros and cons
Pros
- Highest independent overall score of the models we cover (58)
- Leads published coding-agent and office-work tests
- Cheaper and faster than Opus 5
- Available on every major cloud
Cons
- $4 / $20 is double GPT-6.1 Sol and Gemini 4 Argon's launch price
- Some breaking changes for developers coming from Opus 5
- Anthropic's "about 40% cheaper on typical work" held only on some tasks in an independent test
- GPT-6 Astra still leads on AutomationBench and Terminal-Bench Science
How it compares
Opus 5.5 scores 58 on the independent Artificial Analysis index, ahead of Gemini 4 Argon (about 53) and GPT-6.1 Sol, but costs twice as much per token as both. Inside Anthropic's own lineup it beats the pricier Claude Fable 5.1 ($10 / $50) on every test Anthropic compared them on.
Hands-on review In testing
We are running Claude Opus 5.5 on real tasks. Results will be added to this page, at the same address.
How it got here
| Date | What happened | Source type |
|---|---|---|
| Sep 23, 2026 | Independent write-up confirms 1M context and 128K output, and tests the cost claims.ComputingForGeeks, Claude Opus 5.5 features and benchmarks | News report |
| Sep 22, 2026 | Claude Opus 5.5 released at $4 / $20, down from $5 / $25 for Opus 5.OpenRouter, Claude Opus 5.5 model page | Confirmed |
Frequently asked questions
When was Claude Opus 5.5 released?
On 22 September 2026, as the first model of the Claude 5.5 family.
How much does Claude Opus 5.5 cost?
In the API, $4 per million input tokens and $20 per million output tokens, or half that in the Batch API. In the Claude apps it depends on your plan.
Is Claude Opus 5.5 free?
It cannot be downloaded. You use it in the Claude apps, depending on your plan, or pay per token through the API.
What changed from Claude Opus 5?
A 20% lower price per token, more than 30% faster output, thinking that is always on, and higher scores on coding, computer-use and office-work tests.
How much text can Claude Opus 5.5 read?
Up to 1 million tokens in one go, and it can write up to 128,000 tokens in one answer.
Related models
Claude Haiku 5.5
AnthropicAnthropic's small, fast model, 90% cheaper per token than Haiku 4.5.
Claude Fable 5.2
AnthropicPrediction markets expect a Fable point release. Anthropic has not announced anything.
Claude 6
AnthropicThe next major Claude generation. Only prediction markets give a date so far.