Claude Opus 5.5

Cosmic JS's head-to-head comparison notes that Claude Opus 5.5 scores higher than Opus 5 on every benchmark Anthropic published for both models while simultaneously costing less, an unusual combination in recent frontier releases. The model string is claude-opus-5-5, available through the Anthropic API, AWS, Google Clo The pricing table pulled from Anthropic's pricing page on launch day lists Opus 5.5 at $4/$20 input-output per million tokens with $0.20 cache reads and $5 cache writes, compared with Opus 5 at $5/$25, Sonnet 5 at $2/$10, Haiku 4.5 at $1/$5, and Fable 5.1 at $10/$50. The article also flags that Claude Sonnet 5.5 and Cl

LLM Stats ranks Claude Opus 5.5 at the top of the Coding category (1 of 271), Reasoning (1 of 367), and Vision (1 of 211), with Tool Calling at 2 of 198 and Math at 5 of 328, placing it in the "Great" top-2% tier overall. Detailed benchmark scores drawn from Anthropic's system card include a GDPval-AA v2.1 Elo of 1846/ The cost-efficiency chart shows Opus 5.5 at a blended price of $4.76 per million tokens with an LLM Stats Score of 59.8, positioning it well below comparable flagships on the price-per-performance curve. Capability tier rankings show Opus 5.5 at 15 of 246 in Healthcare, 88 of 212 in Legal, and slightly below average on

CodeRabbit's independent evaluation found that Claude Opus 5.5 edged past their production baseline on bug coverage in an open-source test set and made larger gains in both coverage and precision on a smaller set of harder cases, with those gains also bringing more comments and higher reported token usage. The team hig The review also explored Opus 5.5 through an overnight coding project and a Grand Theft Auto: San Andreas-inspired gaming experiment. A notable behavioral change is that thinking is always adaptive in Opus 5.5: the model rejects requests that explicitly enable or disable thinking, so effort becomes the main control ove

Anthropic launched Claude Opus 5.5 on September 22, 2026, positioning it as a new flagship that delivers performance comparable to the company's top-tier Fable 5.1 while costing roughly 40% less to run than its predecessor. The company priced Opus 5.5 at $4 per million input tokens and $20 per million output tokens, a External AI safety research groups Frontier Design and METR tested Opus 5.5 before launch, and Anthropic is expanding programs that give vetted cybersecurity and life-sciences researchers broader access to the model. The release lands amid CEO Dario Amodei's recent public call for the global AI community to slow the pa

Sonar ran Claude Opus 5.5 through their Java benchmark suite covering HumanEval, MBPP, and ComplexCodeEval (4,444 tasks, 544 with executable tests) and recorded an 87.7% pass rate on the executable subset, within one percentage point of Opus 5's 88.6%. Opus 5.5 wrote 27.5% less code (664,890 lines vs Opus 5's 916,813) Severity metrics improved across the board for Opus 5.5: reliability BLOCKER findings fell 41%, security BLOCKER findings fell 53%, and maintainability BLOCKER findings fell 20%. Two counter-metrics moved the other way: bug density rose 12% (from 576 to 644 per mLOC) and concurrency findings rose 44%. Sonar concluded t