Cheaper is the whole pitch

On most work, Anthropic says Opus 5.5 does the job at the level of its stronger Fable 5.1 model while costing about 40% less to run than Opus 5, and it spits out answers more than 30% faster. The price cuts are specific. A million input tokens now run about $4 and a million output tokens about $20, both a fifth cheaper than Opus 5. Cache reads, which the company says eat up most of the bill in coding and agent work, drop 60% to $0.20 per million. It also uses fewer tokens per task, which is where the rest of the saving comes from. You can use it now as claude-opus-5-5 on Anthropic's own platform and through Amazon, Google and Microsoft's clouds, and paying subscribers on Pro, Max, Team and Enterprise plans get higher usage limits on top.

Fast in tests and in real work

On Terminal-Bench 4.0, a coding test, Opus 5.5 scored 66.4%.

That beat Fable 5.1 at 55.8% and OpenAI's GPT-6 Astra at 57.9%, and on a separate benchmark that scores real work across 44 jobs it posted 1846 Elo against 1708 for Opus 5. Anthropic backed the numbers with live runs. One early tester pushed through a 680,000-line code migration in under a day. In an internal test that had both Opus 5.5 and Fable 5.1 rewrite a widely used load balancer from C into Rust, Opus 5.5 finished in 9.5 hours against 12 and cost 51% less. The company also says it writes more plainly now, leading with the point and dropping the jargon, which answers a common gripe about Opus 5.

The safety catch it admits to

Here's the twist. The cheapest Opus in a generation is also the one Anthropic rates as its safest. Outside evaluators including METR and Frontier Design checked it before launch, and the company says it scored the best of any of its models on an automated behaviour audit that runs close to 2,000 scenarios. That safety shows up as friction. It matches the guarded Mythos 5.1 on biology and cybersecurity, most cyber tasks get bumped down to the older Opus 4.8, and vetted labs and security pros have to apply to verification programs for fuller access. Anthropic is honest about the weak spot too, saying Opus 5.5 often senses when it's being tested, which makes its real behaviour harder to predict, and that its own checks may miss things. Cheaper Sonnet and Haiku versions are due in the coming weeks.