Anthropic ships Claude Opus 5.5 and cuts token prices by a fifth
Opus 5.5 charges $4 per million input tokens and $20 per million output tokens, 20% below Opus 5. Cache reads, which dominate the bill for agentic and coding work, fall 60% to $0.20 per million. Anthropic says the model works at the level of Claude Fable 5.1 on most tasks while costing 40% less to run than Opus 5. It also arrives with safeguards that reroute most cybersecurity work to Opus 4.8.
Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in a new Claude 5.5 family. For anyone who pays the bill, the price is the news: input and output tokens cost 20% less than Opus 5, and cache reads cost 60% less. On the Claude Platform the model string is claude-opus-5-5, and Anthropic lists it on Amazon Web Services, Google Cloud, and Microsoft Azure.
What a million tokens costs now
Anthropic publishes four numbers, and the one that moves most is not input or output. Cache reads drop to $0.20 per million tokens from $0.50, a 60% cut, and the company says cache reads make up the majority of what agentic and coding work costs. Input and output each come down 20%.
The headline 40% saving is a different claim, and it rests on two things at once: a lower price per token and fewer tokens spent per task. Anthropic measured that on its own workloads at default settings, so treat it as the vendor's figure until your own traffic says otherwise. A fast mode in Claude Code and the Claude Platform runs at $8 per million input tokens and $40 per million output tokens for up to 2.5 times the speed.
- Input: $4 per million tokens, down from $5.
- Output: $20 per million tokens, down from $25.
- Cache reads: $0.20 per million tokens, down from $0.50.
- Cache writes: $5 per million tokens, down from $6.25.
The scores, and the caveat Anthropic attaches to them
On Terminal-Bench 4.0, Anthropic reports 66.4% for Opus 5.5 against 55.8% for Fable 5.1, 52.3% for Opus 5, 57.9% for GPT-6 Astra, and 37.3% for GPT-5.6 Sol. On GDPval-AA v2.1, which grades professional work across 44 occupations, it reports 1846 Elo against 1735 for Fable 5.1 and 1708 for Opus 5. The Terminal-Bench figure carries a standard error of 2.6 points, so a gap of two or three points there is not a gap.
The company then argues against reading too much into its own table. It says benchmark margins have become a less reliable guide to real-world differences at these capability levels, and that in its own use the distance to Fable 5.1 is narrower than the numbers suggest. A second limitation sits in the footnotes: when the production safeguards intervened during evaluation, cybersecurity tasks were finished by Opus 4.8 and biology tasks by Opus 5, which Anthropic says likely lowered the published scores.
The safeguards decide what the model will do for you
Opus 5.5 is the first Opus model to launch with the class of safeguards Anthropic built for Fable 5.1. You can still find and fix bugs in your own code as part of ordinary development, but most cybersecurity tasks are rerouted to Opus 4.8, a weaker model. Anthropic says a Cyber Verification Program with three tiers of access will expand to cover Opus 5.5, and that verified practitioners will be able to use it for security work.
Biology runs under the same safeguards as Fable 5.1, with a Life Sciences Verification Program that vetted organizations can apply to. A third safeguard targets distillation: preserved thinking blocks API users from editing Claude's earlier context to extract its reasoning, and it applies to API accounts created on or after August 31, 2026. Opus 5.5 is also no longer offered with thinking switched off, and its text output carries a watermark for the EU AI Act.
Where you can reach it today
AWS offers two routes. Amazon Bedrock turns on zero data retention by default, keeps data inside AWS with regional residency, and adds managed pieces such as Guardrails and Knowledge Bases. Claude Platform on AWS gives you the native platform through the AWS console, with the same APIs and console experience, unified with AWS billing and authentication.
GitHub added the model to Copilot the same day for Pro+, Max, Business, and Enterprise plans, billed at provider list pricing under usage-based billing, with a gradual rollout. It appears in the model picker across Visual Studio Code, Visual Studio, Copilot CLI, the Copilot coding agent, github.com, GitHub Mobile, JetBrains IDEs, Xcode, and Eclipse. Anthropic separately raised five-hour usage limits on Pro, Max, Team, and per-seat Enterprise plans, and says Sonnet 5.5 and Haiku 5.5 follow in the coming weeks.
„Benchmark margins have become a less reliable guide to real-world differences.“
Sources
Related
Anthropic releases Claude Sonnet 5.5, with input and output tokens at half the Opus 5.5 price
Claude Sonnet 5.5 keeps Sonnet 5's prices: $2 per million input tokens and $10 per million output tokens. That is half of what Opus 5.5 charges, while Anthropic's own benchmarks put the two models within a few points of each other. Code that turns thinking off needs a change before the switch, because the old setting now returns an error.
Anthropicverified
H Company releases Holo4, open-weight models for agents that operate a computer
H Company released two Holo4 models on September 28, 2026, with weights on Hugging Face. The larger one scores higher on the company's own tests but may not be used commercially. The smaller one is licensed under Apache 2.0.
Hugging Faceverified
OpenAI releases GPT-6 Sol and Luna and halves the API price of both tiers
GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, and GPT-6 Luna costs $0.10 and $0.50. OpenAI measures the 50% cut against the promotional prices it charged for GPT-5.6. The launch also brings a reworked prompt cache with discounts of up to 90% on cached input. Every benchmark comparison with Claude comes from OpenAI and uses Opus 5, not Opus 5.5.
OpenAIverified