Claude Opus 5.5 is Anthropic's current Opus model, released September 22, 2026, built for long-running agentic coding and knowledge work at $4 / $20 per million input / output tokens.

Claude Opus 5.5 is the first model in 📝Anthropic's Claude 5.5 family and the successor to Claude Opus 5 in the 📝Claude Opus tier of the 📝Claude family. Anthropic reports that it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads, and its documentation names it the starting model for most workloads. It generates output tokens more than 30% faster than Opus 5 and tends to finish the same task with fewer tokens. It is also the first Opus model to launch with safeguards similar to Claude Fable 5.1's on cybersecurity, biology, and distillation.

Key Facts

Model ID: claude-opus-5-5 (Amazon Bedrock: anthropic.claude-opus-5-5)
Released: September 22, 2026; retirement not sooner than September 22, 2027
Predecessor: Claude Opus 5, released July 24, 2026
Context window: 1M tokens, 128K max output (300K on the Batch API with a beta header)
Pricing: $4 input / $20 output per million tokens; cached reads $0.20; Batch API $2 / $10
Thinking: adaptive and always on, with a default effort of medium
Knowledge cutoff: June 2026
Status as of: September 28, 2026

Sources: Anthropic — Introducing Claude Opus 5.5, Anthropic — What's new in Claude Opus 5.5.

How It Works

Always-on adaptive thinking — the model decides how much to think; the effort parameter is the only control, and requests that disable thinking or set a manual budget return an error.
Medium default effort — in Anthropic's testing, medium on Opus 5.5 matches or exceeds Opus 5 at high on coding and knowledge-work evaluations.
Long-running agentic coding — sustains multi-hour audits and codebase migrations with parallel subagents, and catches more bugs with fewer false alarms in code review.
Knowledge work — less likely than Opus 5 to state an incorrect figure or cite the wrong source, and stronger at financial models, spreadsheets, and slides.
Visual reading and computer use — reads dense charts more accurately at its lowest effort than Opus 5 did at its highest; computer use requires the computer_toolset_20260801 toolset on the Claude API and 📝Google Cloud.
Progress updates in thinking blocks — notes written between tool calls arrive as thinking blocks, empty unless the client sets thinking.display to "updates".
Safeguard classifiers — cybersecurity, biology, and reasoning-extraction classifiers can decline a request with stop_reason: "refusal", which can be retried on a fallback model.

Why It Matters

Claude Opus 5.5 narrows the gap between the Opus and Mythos tiers. Work that required Claude Fable 5.1 at $10 / $50 per million tokens now runs on an Opus model at 40% of that price, which shifts the default model choice for most agentic and knowledge-work applications.

It also marks the point where Anthropic's Mythos-class safety posture reached the mainstream tier. Opus 5.5 ships with the same class of classifiers as Fable 5.1, so integrations that ran on Opus 5 must now handle refusals and fallback as a normal response path.

FAQ

When was Claude Opus 5.5 released?

Anthropic released Claude Opus 5.5 on September 22, 2026. It is available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, Microsoft Foundry, and Claude Code.

How much does Claude Opus 5.5 cost?

Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, down from Opus 5's $5 and $25. Cached reads cost $0.20 per million tokens, and Batch API requests are half price.

Is Claude Opus 5.5 as good as Claude Fable 5.1?

Anthropic reports that Opus 5.5 performs at Fable 5.1's level on most work. Anthropic recommends Fable 5.1 for the most demanding reasoning and long-horizon agentic work, or when evaluations on Opus 5.5 at higher effort still fall short.

Can you turn off thinking on Claude Opus 5.5?

No. Thinking is always on, and requests that set thinking.type to disabled or enabled return a 400 error. Lowering the effort level is the supported way to reduce thinking, latency, and cost.

How does Claude Opus 5.5 score against Claude Opus 5?

Anthropic reports 66.4% on Terminal-Bench 4.0 versus 52.3% for Opus 5, 57.8% on CursorBench 4.0 versus 46.6%, and 81.8% on OSWorld 2.1 versus 74.0%.

📝How to Prompt Claude Opus 5.5 — symptom-first prompting fixes and operating rules
📝Claude Opus 5 vs Claude Opus 5.5 — what changed from the predecessor
📝Claude Model Glossary — every Claude model by tier and release date
📝Claude Mythos — the tier above Opus that Opus 5.5 approaches
📝Context Engineering Rules for Claude 5 Generation — context practices for the Claude 5 models
📝Claude Code — Anthropic's agentic coding tool, which runs Opus 5.5

We run BotBrian's Claude Code sessions on Opus 5.5 — this memo was researched and drafted in one. Anthropic's own guidance now makes Opus the default starting point again, with Fable 5.1 reserved for the hardest long-horizon work; we will record here how that holds up in practice.