Claude Opus 5.5: Price, Benchmarks and How to Access It

Anthropic released Claude Opus 5.5 on September 22, 2026, and it is available now in the Claude apps and through the API as claude-opus-5-5. The new model costs $4 per million input tokens and $20 per million output tokens, 20% less than Claude Opus 5, and Anthropic says it performs at the level of its pricier Claude Fable 5.1 model on most work while costing about 40% less to run than Opus 5 on typical workloads.

What is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic’s new general-purpose flagship for long-running coding, agent and knowledge work. It replaces Claude Opus 5 as the default Opus model and sits alongside Claude Fable 5.1 and Claude Mythos 5.1 at the top of Anthropic’s lineup. In its official Claude Opus 5.5 announcement, Anthropic describes it as the first model in a new Claude 5.5 family and says it “performs at the level of Claude Fable 5.1 on most work.”

The company’s developer documentation now tells builders to start with Opus 5.5 for most workloads and to move up to Fable 5.1 only for demanding reasoning and long-horizon agentic work, or when Opus 5.5 at higher effort still falls short.

The release is also Anthropic’s first since chief executive Dario Amodei publicly argued for “pacing the frontier,” the idea that AI labs should slow the rate at which they push capabilities forward. Anthropic says the model was tested before release by external evaluators including Frontier Design and METR, and that it ships with the same class of safeguards the company built for its most capable models.

How much does Claude Opus 5.5 cost?

For developers, the headline is a straight price cut. Opus 5.5 is billed at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. Cache reads, which Anthropic notes make up the majority of costs in agentic and coding work, drop from $0.50 to $0.20 per million tokens, a 60% reduction. Cache writes fall from $6.25 to $5.

Anthropic’s claim of a 40% overall saving comes from combining the lower per-token price with the model using fewer tokens per task. The company says Opus 5.5 “requires less compute to serve than Opus 5” and generates output more than 30% faster. A fast mode, available on the Claude API and in Claude Code as a research preview, promises up to 2.5 times the speed at $8 per million input tokens and $40 per million output tokens.

For comparison, the current Claude API price list puts Claude Fable 5.1 at $10 input and $50 output, Claude Sonnet 5 at $2 and $10, and Claude Haiku 4.5 at $1 and $5. Batch processing still takes 50% off, and US-only inference is available at 1.1 times the standard input and output rates.

What about Claude Pro and Max subscribers?

Consumer subscribers do not pay per token, but they do get more room. Alongside the launch, Anthropic says it is increasing five-hour usage limits on the Pro, Max, Team and seat-based Enterprise plans, and giving subscribers a one-off rate limit reset that they can save and use whenever they choose. On Anthropic’s plan comparison, Opus models are included on Pro ($20 a month, or $17 a month billed annually) and on Max 5x and Max 20x, but not on the free tier, which offers Sonnet and Haiku.

How does Opus 5.5 perform? Anthropic’s benchmark numbers

Anthropic published a comparison table against its own previous models and OpenAI’s GPT-6 Astra and GPT-5.6 Sol. All figures below are Anthropic’s reported results (Opus 5.5 at max effort unless noted); some rival scores are taken from OpenAI’s own reports or third-party leaderboards, as Anthropic’s footnotes explain.

The table shows Opus 5.5 ahead on most coding and knowledge-work measures, while GPT-6 Astra keeps a lead on AutomationBench, which Zapier runs, and on Terminal-Bench-Science. Anthropic itself cautions that “benchmark margins have become a less reliable guide to real-world differences” at this level, and says the gap to Fable 5.1 is narrower in its own use than the scores suggest.

The company leans harder on cost-adjusted results. At its default medium effort, Anthropic says Opus 5.5 beats GPT-6 Astra’s top FrontierCode score for about a fifth of the cost per task, matches Astra on Terminal-Bench 4.0 for about 40% of the cost, and beats GPT-5.6 Sol’s best CursorBench score by 11 points for about a third of the cost.

Examples Anthropic highlights

• An early tester completed a 680,000-line code migration in less than a day.
• Asked to cut load times across every page of a web app, Opus 5.5 succeeded 39 of 40 times, according to Anthropic.
• In an internal test translating the HAProxy load balancer from C into Rust, Opus 5.5 finished in 9.5 hours versus 12 for Fable 5.1, at 51% less cost; both passed nearly all of HAProxy’s own regression tests.
• In a research-report test where any invented figure or quote counted as a failure, 16 of 18 Opus 5.5 reports passed Anthropic’s quality bar; neither Fable 5.1 nor Opus 5 passed in any attempt.

Customer testimonials on the launch page follow the same efficiency theme. GitHub’s product chief was among the early testers quoted:

“Developers want agents that can take on real software work and finish it. In our testing across GitHub Copilot CLI and VS Code, Claude Opus 5.5 used among the fewest tokens and steps we measured.”

Claude Opus 5.5 specs at a glance

Anthropic’s models overview documentation lists the technical details developers need to plan a migration.

How to access Claude Opus 5.5

There are three main routes, depending on whether you want a chat app or an API.

1. Claude apps: Subscribers on Pro, Max, Team and Enterprise plans can select Opus 5.5 on the web, desktop and mobile apps, and in Claude Code. The free plan does not include Opus models.
2. Claude API: Developers can call claude-opus-5-5 directly on the Claude Platform; the ID is itself a pinned snapshot, so the alias and the model ID are the same string.
3. Cloud platforms: Anthropic says Opus 5.5 is available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure through Microsoft Foundry. Fast mode is limited to the Claude API.

What developers need to change

Opus 5.5 is not a drop-in swap for every Opus 5 integration. Anthropic’s what’s new in Claude Opus 5.5 guide lists four breaking changes:

• Thinking can’t be disabled. Requests that set thinking to disabled, or that pass a manual token budget, now return a 400 error. The effort parameter is the control for depth, latency and cost.
• Forced tool use returns an error. Setting tool_choice to “any” or to a named tool is rejected; “auto” and “none” still work.
• Thinking blocks are tied to the model and conversation. For API accounts created on or after August 31, 2026, replaying a thinking block after the system prompt, tools or earlier messages have changed returns an error by default. Anthropic calls this “preserved thinking” and describes it as an anti-distillation safeguard.
• The older computer_20251124 tool is not accepted on the Claude API and Google Cloud; computer use now requires the newer toolset. Amazon Bedrock still accepts the old tool.

There are also behavior differences without code changes. The default effort drops from high on Opus 5 to medium on Opus 5.5, the model tends to think more per turn at a given effort level, and short notes between tool calls now arrive as thinking blocks, which can make streaming apps go quiet unless they change a display setting.

Safety testing and new safeguards

Anthropic says Opus 5.5 posted the best scores of any model it has tested on its automated behavioral audit, which covers nearly 2,000 scenarios. In a new evaluation of whether a model tries to cross containment boundaries, the company reports that Opus 5.5 attempted to circumvent boundaries about 85% less often than Opus 5 or Claude Mythos 5.1. It also says Opus 5.5 is more resistant to prompt injection than Opus 5. The full methodology is in the Claude Opus 5.5 system card.

The company is also candid about limits. It says Opus 5.5 “often suspects it is being evaluated,” which makes it harder to know how the model will behave in the wide range of real settings where it is deployed.

Because Anthropic rates Opus 5.5 as comparable to Mythos 5.1 in biology and cybersecurity, it ships with stricter guardrails than earlier Opus models:

• Cybersecurity: Routine bug-finding and fixing in your own code is allowed, but most other cybersecurity tasks are rerouted to Claude Opus 4.8. Anthropic says it will expand its Cyber Verification Program to Opus 5.5 in the coming weeks, with three tiers of increasingly permissive access.
• Biology: Opus 5.5 uses the same biology safeguards as Fable 5.1. Vetted labs, startups and drug companies can apply to the Life Sciences Verification Program for less restrictive access.
• Data and compliance: Opus 5.5 is available with zero data retention and carries Anthropic’s text watermarking, which the company says is part of its EU AI Act compliance.

What happens next

Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow “in the coming weeks” with many of the same performance, efficiency and safety improvements. For now, the practical picture is simple: Opus 5.5 is cheaper per token than the model it replaces, Anthropic claims it needs fewer tokens per task, and it becomes the company’s recommended starting point for most API work.

The launch landed on the same day OpenAI released GPT-6 Sol and GPT-6 Luna with their own price cuts, so developers comparing providers this week are looking at a rapidly shifting cost picture. Anyone upgrading from Opus 5 should budget time to handle the thinking and tool-choice changes and to re-run effort settings, since the default has moved from high to medium.

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