Anthropic's Claude Opus 5.5 launches at $4/$20, 20% below Opus 5

Anthropic released Claude Opus 5.5 (claude-opus-5-5) on September 22, 2026. It costs $4 per million input tokens and $20 per million output tokens, against $5 and $25 for Claude Opus 5, which came out on July 24. It is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry.

What it costs

Per 1M tokens	Opus 5	Opus 5.5
Input	$5	$4
Output	$25	$20
5-minute cache write	$6.25	$5
Cache read	$0.50	$0.20

Cache reads on Opus 5.5 cost 0.05x the input price, half the usual 0.1x. Fast mode, which Anthropic says runs up to 2.5x faster, costs $8 input and $40 output per million tokens. It left research preview on October 15.

Anthropic says that in its own tests, Opus 5.5 costs 40% less than Opus 5 on typical workloads at default settings, and writes output more than 30% faster.

Specs and changes

• 1M-token context window by default and up to 128,000 output tokens.
• Adaptive thinking is always on and can’t be turned off.
• Forced tool use (tool_choice set to any or tool) returns an error.
• The deprecations page says it won’t be retired before September 22, 2027.

How it compares

Anthropic says Opus 5.5 performs at the level of Claude Fable 5.1, which costs $10 and $50, on most work. On its own published scores, Opus 5.5 leads on Terminal-Bench 4.0 with 66.4%, against 55.8% for Fable 5.1 and 57.9% for OpenAI’s GPT-6 Astra, but trails Astra on Terminal-Bench-Science (58.7% vs 64.6%) and AutomationBench (40.0% vs 41.4%). These are vendor numbers.

For Claude app users, the announcement says five-hour usage limits went up on Pro, Max, Team and seat-based Enterprise plans.

What to check

If you run Opus 5 or Fable 5.1 through the API, price a week of your real traffic on Opus 5.5 before switching. Remove forced tool use from your requests first, or they will fail.