Claude Opus 5.5 is here, and Anthropic is playing the price card on its flagship model. On September 24, Anthropic published a technical deep-dive on Claude Opus 5.5 on its official blog: the first model in the Claude 5.5 family, it matches the pricier Claude Fable 5.1 on capability while running about 40% cheaper than Opus 5 on typical workloads. Input and output tokens each cost 20% less, cache reads are 60% cheaper, and output speed is up more than 30%.
The real target of the price cut: the re-reading-context bill
Why cut prices now? Anthropic shared real usage data from Claude Code: from March to September this year, the model work triggered per developer prompt grew 3.3x, context per request grew 2.6x, and the input-to-output token ratio stretched from 189:1 to 324:1. The more capable agents get, the heavier the "re-reading context" part of the bill — a 60% cache-read cut, to $0.20 per million tokens, hits exactly that pain point.
Less fiddling: fewer rounds for the same job
The other selling point is less fiddling: fewer rounds to finish the same task, and over 50% fewer cache-miss inputs. Among early users, an engineer at Clio handed it a large migration spanning six repositories and it ran unattended for more than 18 hours; another 680,000-line code migration test finished in under a day — work Anthropic says used to take engineering teams weeks.
The safety details: the stronger the capability, the higher the guardrails
The notable details are on the safety side. Before release, Opus 5.5 went through third-party evaluations by Frontier Design and METR, with an 85% lower likelihood of attempting to breach its containment boundary than Opus 5. But because its bio and cybersecurity capabilities are close to Claude Mythos 5.1, Anthropic gave this launch its highest tier of safeguards: general cybersecurity prompts are routed to Opus 4.8 instead, and advanced tasks require approved access through the Cyber Verification Program or Life Sciences Verification Program.
The price-war backdrop is worth noting too: the same week, OpenAI's GPT-6 Sol and Luna cut API prices in half. Both companies are calling at the UN Security Council to "pace the AI frontier" while slashing the unit price of flagship capability — efficiency competition is replacing parameter competition as the main event of the second half.
