Anthropic has launched Claude Opus 5.5, kicking off its next generation artificial intelligence lineup. The updated flagship promises 30% faster response times and cuts operating expenses by roughly 40% compared to Opus 5. It is already available across major cloud platforms, with smaller companion models expected to follow soon.
Anthropic adjusted its pricing structure to make large context queries more accessible. Running standard workloads on the new model costs $4 per 1,000,000 input tokens and $20 per 1,000,000 output tokens. Reading cached data runs at $0.20 per 1,000,000 tokens, while writing to the cache costs $5. These prices undercut older Opus 5 rates, which charged $5 for inputs and $25 for outputs alongside higher caching fees.
Raw performance gains show up clearly in internal evaluations. In Terminal Bench 4.0 coding tests, the model achieved 66.4% accuracy, jumping ahead of the 52.3% score posted by its predecessor. The GDPval AA v2.1 evaluation showed a similar lift, climbing to 1846 Elo points from the previous 1708 mark. Anthropic cautioned that test scores do not always reflect real world software tasks, but the model matches the everyday performance profile of Claude Fable 5.1 across standard workflows.
Developers using Claude Code or the official developer platform can toggle a dedicated fast mode. This setting boosts operational speed by up to 2.5 times for latency sensitive jobs. That speed premium raises the pricing tier to $8 per 1,000,000 input tokens and $40 per 1,000,000 output tokens.
Beyond raw speed, the model received styling updates to deliver cleaner output. Anthropic refined the default behavior so answers lead with direct conclusions while cutting back on unnecessary filler words. Independent evaluation groups Frontier Design and METR evaluated the software prior to public deployment. Enterprise teams can access Claude Opus 5.5 immediately through Amazon Web Services, Google Cloud, and Microsoft Azure. Releases for Claude Sonnet 5.5 and Claude Haiku 5.5 will arrive in the coming weeks.
