Claude Opus 5.5

Available · Language, Multimodal, Reasoning

Claude Opus 5.5 is a language model in Anthropic's Claude Opus line and, according to Anthropic, the first model of its Claude 5.5 family, released on 22 September 2026 on all platforms. It costs less per token than Claude Opus 5, has adaptive thinking that cannot be switched off, is available with zero data retention and carries Anthropic's output watermarking. [1] [2]September 22, 2026 Primary source

Timeline of Claude Opus 5.5 →

Claims and evidence

  • Released Primary source[1] [3]Availability: Released
  • Status Available Primary source[4]Compare models table [3]Availability [5]Model status table
  • Change · Other Priced at 4 USD input and 20 USD output per million tokens, down from 5 and 25 USD for Opus 5; cache reads cost 0.20 USD.Compared with Claude Opus 5 Primary source[1] [2]September 22, 2026
  • Change · Reasoning Thinking can no longer be switched off; its depth is set with the effort parameter.Compared with Claude Opus 5 Primary source[2]September 22, 2026
  • Change · Tool use Forced tool choice (tool_choice types any and tool) returns an error.Compared with Claude Opus 5 Primary source[2]September 22, 2026
  • Change · Safety Deployed with safeguards similar to those on Claude Fable 5.1 and with Anthropic's output watermarking.Compared with Claude Opus 5 Primary source[1]
  • Input text, image Primary source[3]Capabilities: Input and output
  • Output text Primary source[3]Capabilities: Input and output
  • Context window 1M tokens Primary source[3]Capabilities: Context window
  • Parameters Not disclosedNot disclosed by Anthropic. No claim madeNo source
  • Access API, consumer app, cloud partner Primary source[1]Availability [3]Availability: Platforms

Lineage

Predecessors

Successors

No known successor.

Based on

Not derived from another model.

Variants and derived

None recorded.

Siblings

None recorded.

All ancestors

All descendants

None.

Variants

No variants recorded in this record.

Related AI Radar coverage

All model releases from Anthropic on AI Radar →