Claude Opus 5.5 is now available in Omni AI. Anthropic released it on September 22, 2026 as the first model of the 5.5 wave, and it is live on the Omni AI Pro plan from today, replacing Claude Opus 5 in the model picker.
This guide covers where to find it, what changed against Opus 5 and the larger Claude Fable 5.1, what the benchmark scores mean, the context window and API pricing, and the limits worth knowing before you move real work onto it.
Claude Opus 5.5 Is Now Available in Omni AI
You do not need a separate Anthropic subscription to use it. Opus 5.5 sits in the same model picker as GPT-6 Astra, Gemini, Grok and DeepSeek, on one account and one subscription.
- Where to find it: the Omni AI Pro plan, on iPhone, Android and the web, in the model picker on any chat.
- What changed: Claude Opus 5.5 replaced Claude Opus 5, which is no longer listed. Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5 are unchanged.
- Your old chats: conversations that used Opus 5 keep working and are served by 5.5 automatically, so nothing needs to be re-created.
- One subscription: the same Pro plan covers every model in the list, so switching between them costs nothing extra.
How to Use Claude Opus 5.5 in Omni AI
Four steps, and the model is chosen per message rather than per account:
- Open Omni AI on iPhone or Android, or use it in a browser without installing anything.
- Start a chat and open the model picker at the top of the conversation.
- Choose Claude Opus 5.5 from the Anthropic models. It is part of the Pro plan.
- Switch whenever you like. You can change model in the middle of a conversation and the new one sees the whole thread, which is the point of keeping several in one app.
What Is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's flagship model, built for agentic coding and knowledge work: long tool-using sessions, operating software, editing real repositories and producing finished professional output rather than a single good answer. Anthropic describes it as the strongest model it has tested to date, and says it performs at the level of Claude Fable 5.1 on most work despite being the cheaper model to serve.
The efficiency story is the interesting part. Opus 5.5 generates output roughly 30% faster than Opus 5 and uses fewer tokens to finish the same task, which is why Anthropic puts the saving on a typical workload at about 40% rather than quoting the per-token price alone. Anthropic has said Claude Sonnet 5.5 and Claude Haiku 5.5 follow in the coming weeks, so this is the start of a generation rather than a one-off release.
Claude Opus 5.5 Benchmarks: What the Numbers Actually Say
The pattern in Anthropic's published scores is consistent rather than lopsided. Opus 5.5 leads both Fable 5.1 and Opus 5 on every headline benchmark, and the margin is widest on agentic coding.
- Terminal-Bench 4.0: 66.4% against 55.8% for Fable 5.1 and 52.3% for Opus 5. This is the largest gap in the set, and it measures real terminal work rather than snippets.
- FrontierCode v1.1: 54.4% against 50.3% and 48.0%. A clear lead, and a reminder that half of these tasks still fail.
- CursorBench 4.0: 57.8% against 51.8% and 46.6%, on editing inside real codebases.
- GDPval-AA v2.1: 1846 Elo against 1735 and 1708, on economically valuable knowledge work rather than puzzles.
- OSWorld 2.0: 81.8% on computer use, reported as a partial result.
- Humanity's Last Exam, with tools: 67.7%, the broad open-domain reasoning test where frontier models have been trading places all year.
Read together, the takeaway is narrower than the headline. If your work is agentic coding, operating software, or long professional deliverables, Opus 5.5 is the strongest option on published numbers today. These are vendor-run evaluations, so the usual discount applies until independent testing catches up.
Claude Opus 5.5 vs Fable 5.1 vs Opus 5
Fable 5.1 was Anthropic's largest model and Opus 5 was the previous flagship, so both are fair reference points. Four differences matter more than the rest:
- Agentic coding: a 10.6 point lead over Fable 5.1 on Terminal-Bench 4.0, which is a wider gap than most full generation jumps.
- Price against Fable: $4 and $20 per million input and output tokens, against $10 and $50 for Fable 5.1. The smaller model is both cheaper and ahead on these tests.
- Cost against Opus 5: Anthropic puts a typical workload at about 40% less, because the model uses fewer tokens as well as costing less per token.
- Speed: around 30% faster output generation than Opus 5, which changes how much of a long agentic run you are willing to sit through.
Context Window and Pricing
The context window is unchanged at the top end of the market, and the pricing is where this release actually moves. These are Anthropic's API rates; inside Omni AI the model is part of a flat subscription with no per-token billing.
- Context window: 1 million input tokens, roughly 1,500 pages of text.
- Output limit: up to 128,000 output tokens in a synchronous response, and up to 300,000 through the batch API.
- Standard API pricing: $4 per million input tokens and $20 per million output tokens.
- Caching: $0.20 per million tokens on cache reads and $5 on cache writes, with reads around 60% cheaper than on Opus 5. Caching is the single biggest lever on a long agentic run.
- Fast mode: $8 input and $40 output per million tokens, for when latency matters more than cost.
What Claude Opus 5.5 Is Best For
Every model has a shape. Opus 5.5 is built for work that runs long, touches tools, and has to be finished rather than merely started:
- Agentic coding: multi-step work in a real repository, where the model runs commands, reads output and corrects itself instead of returning one block of code.
- Terminal and computer use: operating software across many steps, which is where its largest benchmark lead sits.
- Professional knowledge work: the kind of economically valuable tasks GDPval measures, such as analysis, drafting and review against a real brief.
- Long-context research: holding a million tokens of source material together well enough to be worth trusting.
- Writing that has to stay clear: Anthropic says it cut jargon and puts the important information at the start of a response.
A frontier model is no longer judged on whether the answer sounds smart. It is judged on whether the task is finished, correct, and done for less than the last model charged.
Limitations and What to Watch
Anthropic reports that Opus 5.5 is the strongest performer on its automated behavioural audit, the most comprehensive alignment test it runs, while stating plainly that evaluating alignment remains an unsolved problem. Both halves of that sentence are worth keeping. The model also ships with the safeguards Anthropic applies to its most capable models, which restrict work on biological and cyber offensive capability.
Two practical caveats matter more day to day. Anthropic applies a preserved thinking safeguard to API accounts created after August 31, 2026, which limits how much of the model's reasoning trace you can see, so a workflow built on reading chain of thought may behave differently on a new account than on an old one. And every number above is published by the vendor, on benchmarks the vendor selected. The scores are plausible and the direction is consistent, but a 10 point lead on a vendor-run test is not the same as a 10 point lead on your workload.
No single model wins everything, which is the reason to keep several within reach rather than commit to one. The benchmarks above show Opus 5.5 ahead on agentic coding while the open-domain frontier stays contested, so being able to put the same question to two models and keep the better answer is worth more than chasing whichever name is newest. For the longer version of that argument, read which AI model is actually the smartest, and for the other side of this month's frontier race, see our breakdown of GPT-6 Astra.
Frequently Asked Questions
Is Claude Opus 5.5 available in Omni AI?
Yes. Claude Opus 5.5 is live on the Omni AI Pro plan, on iPhone, Android and the web. It replaced Claude Opus 5 in the model picker, and sits alongside Claude Opus 4.8, Claude Sonnet 5, Claude Haiku 4.5 and models from OpenAI, Google, xAI and DeepSeek on one account.
How do I use Claude Opus 5.5 on iPhone or Android?
Open Omni AI, start a chat, open the model picker at the top of the conversation and choose Claude Opus 5.5. It is part of the Pro plan. You can switch models in the middle of a conversation and the new model sees the whole thread.
What happened to Claude Opus 5?
Claude Opus 5.5 replaced it. Opus 5 is no longer listed in the model picker, and existing conversations that used it are served by 5.5 automatically, so saved chats keep working without any action from you.
Is Claude Opus 5.5 free?
Not from Anthropic, which lists it on the Pro, Max, Team and Enterprise plans and through the paid API. In Omni AI it is part of the Pro plan, though the app itself is free to download and includes free models you can use without subscribing.
What is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's flagship AI model, released on September 22, 2026 as the successor to Claude Opus 5. It is built for agentic coding and knowledge work, has a 1 million token context window, and is the first model in Anthropic's 5.5 generation.
How much does Claude Opus 5.5 cost in the API?
$4 per million input tokens and $20 per million output tokens. Cache reads are $0.20 per million and cache writes $5, while fast mode costs $8 input and $40 output per million tokens. Anthropic puts the total saving against Opus 5 at roughly 40% on a typical workload, because the model also uses fewer tokens per task.
Is Claude Opus 5.5 better than Claude Fable 5.1?
On Anthropic's published benchmarks, yes, and at a lower price. Opus 5.5 scores 66.4% against 55.8% on Terminal-Bench 4.0, 54.4% against 50.3% on FrontierCode v1.1, 57.8% against 51.8% on CursorBench 4.0, and 1846 Elo against 1735 on GDPval-AA v2.1, while costing $4 and $20 per million tokens against Fable's $10 and $50.
What is the Claude Opus 5.5 context window?
1 million input tokens, with up to 128,000 output tokens in a synchronous response and up to 300,000 through the batch API.