Claude Opus 5.5 Arrives in Claude Code
Claude CodeAnthropic shipped Claude Opus 5.5 on September 22, 2026, and made it the new default Opus model inside Claude Code (release 2.1.280). The model carries a 1 million token context window, cuts pricing to $4 per million input tokens and $20 per million output tokens with cache reads down to $0.20 per million, and Anthropic says it matches Claude Fable 5.1 on most work while costing 40 percent less to run than Opus 5. Independent benchmarks show Opus 5.5 completing agentic coding and terminal tasks in fewer steps and tokens than its predecessor, and Anthropic's own blog post on task economics frames it as the recommended daily driver for supervised coding work in Claude Code.
Featured Video
A video we selected to help illustrate this changelog
Key Takeaways
- Code review catches more, at a cost: independent tests show Opus 5.5 finds more real bugs than Opus 5 in harder cases, but also produces more comments and higher token usage doing it.
- Less code for the same result: Opus 5.5 writes about 27.5% less code than Opus 5 for equivalent tasks while holding the same pass rate, with fewer code-smell and vulnerability findings.
- Fewer steps to the same answer: in VS Code and terminal benchmarks, Opus 5.5 solved more tasks than Opus 5 using less than half the steps at matched effort levels.
- Plainer status updates: the model reports on its work more directly, leading with what it did and what it needs from the user instead of burying it in jargon.
- A 1M-token context window at a lower price: Claude Code's new default Opus model runs $4/$20 per million input/output tokens, with cache reads cut 60 percent to $0.20 per million.
- Anthropic's first release since its "pace the frontier" pledge: Opus 5.5 launches after CEO Dario Amodei said Anthropic would slow capability gains to match alignment progress, and Anthropic says its own behavioral audit found Opus 5.5 to be its best-behaved model yet.
Sources & Mentions
5 external resources covering this update
Claude Opus 5.5 for Code Review: More Catches, More Tokens
YouTube
Anthropic releases Opus 5.5 with lower prices and Fable-level performance
TechCrunch
Anthropic Launches Claude Opus 5.5 With Fable-Level Performance at a Lower Price
MacRumors
Anthropic upgrades Claude with new Opus 5.5 model, details here
9to5Mac
Claude Opus 5.5 Code Review Benchmarks
CodeRabbit Blog
A New Default Opus Model
Claude Code 2.1.280, released September 22, 2026, added Claude Opus 5.5 (claude-opus-5-5) and made it the default Opus model across the terminal, the VS Code extension, Claude Code on the web, and Cowork. Opus 5.5 is the first model in Anthropic's new 5.5 family and Anthropic's first release since CEO Dario Amodei said the company would pace its capability gains to match its progress on alignment. Anthropic reports that on its automated behavioral audit, Opus 5.5 is the best-behaved model it has tested, and the model was reviewed before release by external evaluators including Frontier Design and METR.
Pricing and Context
Opus 5.5 ships with a 1 million token context window and meaningfully cheaper pricing than Opus 5: $4 per million input tokens and $20 per million output tokens, both a 20 percent cut, plus cache reads down 60 percent to $0.20 per million and cache writes at $5 per million. A fast mode is also available, at $8/$40 per million input/output tokens, for roughly 2.5x faster output. Anthropic frames the combination as a 40 percent cost reduction versus Opus 5 on typical workloads, alongside output generation more than 30 percent faster than its predecessor.
Benchmarks and What Anthropic Says About It
Anthropic's own published numbers put Opus 5.5 ahead of both Opus 5 and Claude Fable 5.1 across agentic coding and reasoning evaluations, including Terminal-Bench 4.0 (66.4% vs. 52.3% for Opus 5), FrontierCode v1.1 (54.4% vs. 48.0%), CursorBench 4.0 (57.8% vs. 46.6%), and OSWorld 2.0 (81.8% vs. 74.0%). Independent testing found similar results: in VS Code and terminal benchmarks, Opus 5.5 solved more tasks than Opus 5 while using less than half the steps at matched effort, and at medium effort it matched or beat Opus 5's high-effort results using roughly half the tokens.
Code-review-specific testing found a real tradeoff: Opus 5.5 catches more genuine bugs than Opus 5, particularly in harder cases, but does so with more comments and higher token usage. It also writes about 27.5 percent less code than Opus 5 for equivalent tasks while holding the same pass rate, and produces fewer code-smell and vulnerability findings across the board.
What It Means for Claude Code Users
Anthropic's companion blog post on task economics positions Opus 5.5 as the recommended "daily driver" for supervised coding work in Claude Code, best suited to feature work across a few files, debugging, and code review with follow-up edits, while suggesting smaller models stay in rotation for simple lookups and searches. The post also notes that cache-hit rate is now the single biggest lever on session cost: a session that is mostly cache reads can save up to 60 percent on input spend, and raising effort adds roughly 20,000 thinking tokens per task (about $0.40 on Opus 5.5), which is often cheaper than a single failed retry.