Anthropic introduced Claude Opus 5.5, the first model in the Claude 5.5 family. Official framing: it performs at the level of Claude Fable 5.1 on most work and costs about 40% less to run than Opus 5 on typical workloads. It is the first Anthropic release since CEO Dario Amodei’s call to “pace the frontier,” and it shipped with external pre-release evals from Frontier Design and METR plus the strongest automated behavioral audit score Anthropic says it has tested.
The commercial move is clear: lower list prices than Opus 5 (especially cache reads), faster default output, an optional Fast mode, and broader cloud availability — Claude Platform, AWS, Google Cloud, and Azure — under model id claude-opus-5-5. Sonnet 5.5 and Haiku 5.5 are promised in the coming weeks.
Image credit: Anthropic
Key points
- What shipped: Claude Opus 5.5 (
claude-opus-5-5) — Fable-class capability on most work at ~40% lower typical run cost than Opus 5; first Claude 5.5 family model. - Where: Claude Platform, Amazon Web Services, Google Cloud, Microsoft Azure.
- Pricing (per 1M tokens vs Opus 5): Input $4 (was $5) · Output $20 (was $25) · Cache reads $0.20 (was $0.50, 60% less) · Cache writes $5 (was $6.25).
- Speed: Output more than 30% faster than Opus 5 at default settings.
- Vendor benches (self-reported; adaptive/max effort notes apply): Terminal-Bench 4.0 66.4% · FrontierCode v1.1 54.4% · CursorBench 4.0 57.8% · GDPval-AA v2.1 Elo 1846 (vs Fable 1735 / Opus 5 1708).
- Safeguards: Fable-like cyber + biology controls (cyber often re-routes to Opus 4.8; biology to Opus 5); Life Sciences Verification Program; Cyber Verification Program expanding; preserved thinking anti-distillation for API accounts created on/after Aug 31 2026; zero data retention available; EU AI Act watermarking; thinking mode cannot be switched off.
- Same-day counter: OpenAI shipped GPT-6 Sol and Luna ~90 minutes later (~50% cheaper than the GPT-5.6 Sol/Luna series). Both drops sat near #1/#2 on HN (~1240–1260 points each) within ~24h.
- Caveats: Scores are vendor-reported and often effort-/harness-coupled; production safeguards can hand cyber/biology tasks to older models mid-eval. 9to6AI has not hands-on audited coding quality.
What shipped
Anthropic positions Opus 5.5 as a major step up from Opus 5 on complex coding, agentic work, and knowledge work, while narrowing the real-world gap to Fable 5.1 more than some headline benches imply. Early-tester anecdotes in the launch post emphasize long migrations, overnight multi-repo jobs, fewer steps/tokens per task, and clearer writing.
