Anthropic ships Opus 5.5: Fable 5.1-class work at 40% lower cost

Anthropic has launched Claude Opus 5.5, the first model in its 5.5 family, positioning it as a new frontier leader that matches the more expensive Fable 5.1 on most tasks while running roughly 40% cheaper than Opus 5. The savings come from two directions: input and output tokens drop 20% (to $4 and $20 per million), cache reads fall 60% (to $0.20 per million), and the model burns fewer tokens per task. Output is also more than 30% faster, and subscription tiers get higher usage limits. An optional fast mode trades price for up to 2.5x speed at $8/$40 per million tokens.

The headline gains are in long-horizon agentic coding. Anthropic cites testers completing a 680,000-line migration in under a day and auditing a 200,000-line codebase in under three hours—work that took Opus 5 over 20 hours and 2.5x the tokens. In an internal C-to-Rust port of HAProxy, Opus 5.5 finished faster and 51% cheaper than Fable 5.1 while passing nearly all regression tests. The company claims it beats GPT-6 Astra and GPT-5.6 Sol on several coding benchmarks at a fraction of the cost, though it concedes benchmark margins are now a weak proxy for real-world differences.

Security and alignment are pitched as the differentiator. Opus 5.5 posts Anthropic’s best-ever scores on its automated behavioral audit, is more resistant to prompt injection (tying Fable 5.1 for the lowest attack success rate in Gray Swan testing), and ships with an action-screening classifier, an auditable open-source sandbox, and pre-merge code review. Because its biology and cyber capabilities approach Mythos 5.1 levels, Anthropic is gating those uses behind verification programs for vetted life-sciences and cybersecurity practitioners. Sonnet 5.5 and Haiku 5.5 are due in the coming weeks.