Anthropic says Claude Opus 5.5 costs about 40% less to run
That is versus Opus 5 on typical workloads; the first Claude 5.5 model runs on AWS, Google Cloud and Azure, Anthropic says.
Share
Anthropic shipped Claude Opus 5.5, the first model in its Claude 5.5 family — live on Claude Platform (claude-opus-5-5), AWS, Google Cloud, and Azure. Company pitch: Fable 5.1–class work on most tasks, ~40% lower run cost than Opus 5 on typical workloads, stronger alignment audit scores, and external pre-release testing by Frontier Design and METR.
Price / speed: Input $4 and output $20 per million tokens (20% under Opus 5’s $5/$25). Cache reads $0.20 per million (60% under Opus 5’s $0.50) — the lever for agentic/coding bills. Anthropic: default settings net ~40% lower cost vs Opus 5; output >30% faster. Fast mode in Claude Code / Platform: up to 2.5× speed at $8/$40. Pro/Max/Team/Enterprise five-hour usage limits up; subscription users get a savable rate-limit reset.
What it does better (company): Agentic coding, computer use, knowledge work. Headline benches (attribute to Anthropic; effort footnotes apply): Terminal-Bench 4.0 66.4%; FrontierCode v1.1 54.4%; CursorBench 4.0 57.8%; GDPval-AA v2.1 1846; AutomationBench 40.0% (Zapier; safeguard fails counted as fails). Early-tester framing: long migrations/audits, fewer tokens/steps vs Opus 5; clearer writing (“puts the most important information up front”).
Safety / gates: Best Anthropic automated behavioral-audit scores to date; less hard-to-reverse action / prompt injection vs recent models. Cyber + bio capability treated like Fable 5.1: most cyber tasks re-route to Opus 4.8; biology uses Fable-class safeguards; vetted orgs apply to Life Sciences Verification Program; Cyber Verification expanding. Preserved thinking anti-distillation (API accounts from Aug 31, 2026+). Zero data retention option; watermarking for EU AI Act; thinking mode cannot be switched off. Sonnet 5.5 and Haiku 5.5 “coming weeks.”
Who for: Teams running long agentic coding and knowledge-work sessions who want Opus-class quality at lower token + cache cost. Replaces “stay on Opus 5 / jump to Fable” only if the cost-performance curve matches their harness.
Risks: Vendor benches and customer quotes rule the scoreboard until third parties reproduce. Safeguard fallbacks can mute cyber/bio scores and workflows. “Pacing the frontier” framing is company policy narrative.
Source: Anthropic, Claude Opus 5.5, Sep 22, 2026.
— Rowan Keel
Edition Sept. 22, 2026 · Archive shelf
