Anthropic launches Claude Opus 5.5 — Fable 5.1-level performance at ~40% lower cost, with cyber/bio safeguards that reroute risky requests

Anthropic released Claude Opus 5.5 on Sept 22, 2026, its first Claude 5.5 model, saying it matches Fable 5.1 on most work at about 40% lower operating cost than Opus 5 ($4/$20 per million tokens), while rerouting cybersecurity tasks to Opus 4.8 and flagged…

Anthropic released Claude Opus 5.5 on Tuesday, September 22, 2026, the first model in the Claude 5.5 family and its first release since CEO Dario Amodei pledged to slow the pace of frontier development. The company says the closed-weight API model (claude-opus-5-5, on the Claude Platform, AWS, Google Cloud, and Azure) matches Claude Fable 5.1 on most work, outpaces the larger Fable on many coding and knowledge-work benchmarks, generates output more than 30% faster, and costs about 40% less to run than Opus 5. List pricing fell 20% to $4 per million input tokens and $20 per million output tokens (from $5/$25), with cache reads at $0.20 and a 2.5x fast mode at $8/$40. Because its biology and cybersecurity capabilities are comparable to Mythos 5.1, Fable-class safeguards apply: cybersecurity requests are rerouted to Opus 4.8 and flagged biology requests to Opus 5. New protections include 'preserved thinking' (always-on reasoning that blocks context tampering and distillation attempts), a coding-agent action classifier, an open-source sandbox, roughly 85% fewer containment-boundary circumvention attempts than Opus 5 and Mythos 5.1, Anthropic's best automated behavioral-audit score, and EU AI Act watermarking. METR and Frontier Design evaluated the model before release; Sonnet 5.5 and Haiku 5.5 are expected in the coming weeks. OpenAI shipped GPT-6 Sol and Luna about an hour later the same day. Elsewhere for Anthropic: earlier on Sept 22 the status page showed elevated errors on Claude Opus 5 (with Fable 5/5.1 and Mythos 5/5.1 recovered and the cause identified), and on Sept 23 the company detailed a two-week sprint that made core claude.ai and the desktop app about 3x faster.

• Released Tuesday, Sept 22, 2026; first model in the Claude 5.5 family and first Anthropic release since Dario Amodei pledged to pace frontier development.
• Anthropic says it matches Claude Fable 5.1 on most work, beats the larger Fable on many coding and knowledge-work benchmarks, and runs about 40% cheaper than Opus 5 with output more than 30% faster.
• Pricing: $4/M input and $20/M output (down from $5/$25 for Opus 5), cache reads $0.20/M; a fast mode offers 2.5x speed for $8/$40.
• Closed-weight API model ID claude-opus-5-5 on the Claude Platform, AWS, Google Cloud, and Azure; now the default in Claude Code and the Claude app at medium effort.
• Reported scores: Terminal-Bench 4.0 66.4% (vs 55.8% for Fable 5.1 and 57.9% for GPT-6 Astra), FrontierCode v1.1 54.4%, FrontierCode 1.1 Extended 65.3%, CursorBench 57.8%, OSWorld 2.0 81.8%; GPT-6 Astra still leads Terminal-Bench-Science…
• Safeguards: cyber and biology capabilities comparable to Mythos 5.1 trigger Fable-class safeguards; cybersecurity requests reroute to Opus 4.8 and flagged biology requests to Opus 5 (Help Net Security says most cyber tasks reroute to Opus…
• Prompt-injection defenses: better resistance than Opus 5, about 85% fewer containment-boundary circumvention attempts than Opus 5 and Mythos 5.1, plus a coding-agent action classifier and an open-source sandbox.
• Preserved thinking: reasoning cannot be disabled and blocks API users from tampering with prior context or distilling the model (restricted on new API accounts).