Claude Opus 5.5 Explained: Price, Release Date, Access

Claude Opus 5.5 is Anthropic's flagship model for long-running agentic coding and knowledge work, released on September 22, 2026, priced at $4 per million input tokens and $20 per million output tokens through the API, and available immediately in the Claude apps, Claude Code, the Claude API, and through Amazon Bedrock, Google Cloud, and Microsoft Foundry — but not on Anthropic's free Claude.ai plan.

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's updated Opus-tier model, built for long-running agentic coding and knowledge work. It sits at the top of Anthropic's current lineup alongside Claude Fable 5.1, Claude Sonnet 5.5, and Claude Haiku 5.5, and its model page describes it as the model to start with "for most workloads," with Fable 5.1 reserved for the most demanding reasoning and long-horizon agentic tasks. On Anthropic's own comparative-latency scale it sits in the "moderate" tier — slower than Sonnet 5.5 and Haiku 5.5, but faster than Fable 5.1.

It replaces Claude Opus 5 as Anthropic's primary high-capability model, carrying the API identifier claude-opus-5-5 across the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

Like the rest of the current Claude lineup, Opus 5.5 accepts both text and image input and returns text output, supports multilingual use, vision, and tool use, and runs with "adaptive thinking" that decides on its own how much internal reasoning a given prompt needs. Developers steer that behavior with an effort parameter (low, medium, or high) rather than a separate thinking toggle — Opus 5.5 defaults to `medium` effort, which Anthropic uses as the baseline for most of its own published benchmark numbers. In practice, that means the same prompt can be run cheaper and faster at lower effort, or pushed harder at high effort for problems that need it, without switching models.

When was Claude Opus 5.5 released?

Anthropic released Claude Opus 5.5 on September 22, 2026, according to the model's own specification page, which also lists a retirement commitment of "not sooner than September 22, 2027" for Anthropic-operated platforms (Amazon Bedrock and Google Cloud set their own retirement dates independently). Anthropic has said Claude Sonnet 5.5 and Claude Haiku 5.5 — the rest of the 5.5 lineup — would follow in the weeks after the Opus 5.5 launch, and both are now listed as shipped models in Anthropic's model-comparison table.

Claude Opus 5.5 pricing: the API

On the Claude API, Claude Opus 5.5 is billed per million tokens, with separate rates for cached and batch requests. It is priced lower than Claude Opus 5 on both input and output tokens, and Anthropic's own announcement lists it as 40% cheaper than Opus 5 for typical workloads at default effort settings. Full details, including per-platform variations, are in Anthropic's official pricing documentation.

For comparison, Claude Sonnet 5.5 — Anthropic's mid-tier model in the same generation — is priced at $2 input / $10 output per MTok, roughly half of Opus 5.5's rate. If your workload doesn't need Opus-level capability, it's worth checking what Claude Sonnet 5.5 costs and what it's built for before committing API budget to the larger model.

Is Claude Opus 5.5 free? Claude.ai and Cowork plan pricing

No — Claude Opus 5.5 is not available on Anthropic's free Claude.ai plan. Anthropic's pricing page confirms the free tier has no Opus access at all; to use Opus 5.5 in the Claude apps (including Claude Code and Cowork workflows), you need a paid plan.

Anthropic notes that prices and plan structures are subject to change, and that usage limits apply even on paid plans — Anthropic has said it increased five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans around the Opus 5.5 launch, and gave subscription users a rate-limit reset they can save and use later.

Context window and output limits

Claude Opus 5.5 accepts text and image input and produces text output, with a context window of 1 million tokens — roughly 555,000 words on Anthropic's current tokenizer. Standard output through the synchronous Messages API is capped at 128,000 tokens; developers using the Message Batches API can request up to 300,000 output tokens with the output-300k-2026-03-24 beta header. The model's reliable knowledge cutoff and training data cutoff are both listed as June 2026.

One behavioral change worth knowing if you're migrating from Opus 5: thinking can no longer be switched off on Opus 5.5. It runs with adaptive thinking "always on," defaulting to `medium` effort, and depth is controlled through the effort parameter rather than a thinking toggle — see Anthropic's what's new in Claude Opus 5.5 guide for the full list of breaking changes.

Key capabilities and benchmarks

Anthropic positions Claude Opus 5.5 as performing close to Claude Fable 5.1 — its most capable model — on most work, while costing substantially less to run. Anthropic's own benchmark figures, published alongside the model, include the following comparisons against Claude Opus 5 and Claude Fable 5.1:

Two of the benchmarks above — Terminal-Bench 4.0 and OSWorld 2.1 — test the kind of long-running, tool-using agentic work Anthropic says Opus 5.5 is built for: operating a computer or terminal across many steps toward a goal, rather than answering a single isolated question. FrontierCode and CursorBench are aimed more directly at software engineering tasks, while GDPval-AA and Humanity's Last Exam are broader knowledge-and-reasoning evaluations. Anthropic reports these figures at its default (`medium`) effort setting for Opus 5.5 unless stated otherwise, and notes that scores at higher effort levels can differ.

Anthropic also reports that Opus 5.5 generates output more than 30% faster than Opus 5, and cautions in its own materials that benchmark margins become a less reliable guide to real-world differences at this capability level — a point worth keeping in mind before treating any single score as decisive. We have not independently reproduced these benchmark figures; they are quoted directly from Anthropic's published materials, and third-party benchmark sites may report different numbers under different test conditions.

How Claude Opus 5.5 compares to other Claude models

Within Anthropic's current lineup, Claude Opus 5.5 is positioned as the general-purpose choice for long-running agentic coding and knowledge work, between Claude Fable 5.1 (for the hardest reasoning and longest-horizon agentic tasks) and Claude Sonnet 5.5 (for faster, cheaper everyday use). All four current-generation models — Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 — now share a 1-million-token context window and a 128K-token standard output limit, with pricing as the main differentiator.

Because Opus 5.5 is a direct upgrade rather than a different product line, most comparisons you'll see are against its own predecessor or against Anthropic's other 5.5-era models rather than external competitors; for a head-to-head against a specific alternative model, see our comparison of Claude Opus 5.5 and Claude Fable 5.1.

In practical terms, the choice between these models usually comes down to budget and task shape rather than raw capability. A high-volume classification or extraction pipeline is likely better served by Haiku 5.5's lower per-token cost; an interactive coding assistant or document-drafting workflow that needs strong quality at reasonable latency is the use case Anthropic built Sonnet 5.5 and Opus 5.5 for; and the hardest, most open-ended agentic tasks — the kind that run for a long time with many tool calls and little human supervision — are where Anthropic still points developers toward Fable 5.1 or Opus 5.5 at higher effort settings.

Safety and guardrails

Anthropic's materials describe several safety-related changes alongside the capability and pricing updates. The company reports that Opus 5.5 attempted to circumvent containment boundaries roughly 85% less often than Opus 5 in internal testing, and that it is more resistant to prompt-injection attacks than its predecessor. Cybersecurity-related safeguards mirror those applied to Fable 5.1: most cybersecurity tasks are routed to Claude Opus 4.8 rather than handled directly by Opus 5.5, while routine software bug-fixing work remains allowed. Biology-related safeguards also mirror Fable 5.1's, with expanded access available to vetted organizations through Anthropic's Life Sciences Verification Program. Full detail on the evaluations behind these decisions is published in Anthropic's Claude Opus 5.5 system card.

Anthropic frames this kind of safety work as part of a broader commitment that extends beyond any single model release — including structural commitments like the Anthropic Public Benefit Fund, which the company set up separately from its model safety program.

How to access Claude Opus 5.5

You can reach Claude Opus 5.5 through several channels:

• Claude apps: Available to Pro, Max, Team, and Enterprise subscribers in Claude.ai, Claude Code, and Cowork workflows. Not available on the free plan.
• Claude API: Call it directly with the model ID claude-opus-5-5, billed at the rates above, including zero-data-retention options for eligible accounts.
• Cloud platforms: Available on Amazon Bedrock (anthropic.claude-opus-5-5), Google Cloud Vertex AI, and Microsoft Foundry, each using their own regional availability and billing.

A few technical details matter if you're building against the API directly. The minimum cacheable prompt length is 512 tokens, so very short prompts won't benefit from prompt caching's discounted read rate. Fast mode — a lower-latency version of Opus 5.5 priced separately from the standard rate — is available on the Claude API and Claude Platform for workloads where response speed matters more than per-token cost. And organizations with zero-data-retention agreements in place can use Opus 5.5 under those same terms, with no change in process from earlier Claude models.

What's next for Claude Opus 5.5

Claude Opus 5.5 is Anthropic's current top-of-lineup Opus-tier model, with Claude Sonnet 5.5 and Claude Haiku 5.5 now shipped alongside it as the rest of the 5.5 generation. Anthropic's retirement commitment keeps Opus 5.5 active on its own platforms through at least September 22, 2027, though Amazon Bedrock and Google Cloud can set independent retirement schedules. Developers migrating existing Opus 5 integrations should check Anthropic's migration guide before switching, since thinking can no longer be disabled, forced tool use is no longer supported, and the older computer_20251124 computer-use tool is not accepted on the new model. For most other users, the practical takeaway is simpler: Opus 5.5 is cheaper per token than its predecessor, runs on the same paid Claude.ai plans as before, and is worth testing against your existing Opus 5 workloads given the price cut and reported capability gains.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage