Anthropic began shipping Claude Opus 5.5 this week, slotting it into the lineup as a cost-conscious alternative that closes most of the distance to the pricier Claude Fable 5.1 while cutting the outgoing Opus 5's typical running costs by about 40% at default settings. The release fits a broader pattern among frontier AI labs, where price and efficiency are increasingly the sharper competitive lever now that raw benchmark leads have narrowed; Anthropic itself frames the Fable 5.1 comparison that way, saying the real gap between the two models is narrower in its own use than benchmark scores alone suggest.
Anthropic's Claude Opus 5.5 announcement lays out the mechanics behind that discount. List prices for input and output tokens dropped by roughly a fifth, to $4 and $20 per million respectively, but Anthropic cut the price of reused context far more aggressively, since it says those cached tokens account for most of what an agentic or coding workload actually spends: cache reads fell to $0.20 per million from $0.50 and cache writes to $5 from $6.25. Developers willing to pay a premium can also opt into a faster mode inside Claude Code and the Claude Platform, one that Anthropic says runs up to 2.5 times quicker, priced at double the standard rate.
On its own benchmark suite, Anthropic said Claude Opus 5.5 tops rival models on agentic coding, computer use, and knowledge work, though it cautioned that at this level of capability, score gaps have become a less dependable proxy for real-world differences. The company separately claimed Opus 5.5 posted its best-ever internal alignment audit results and holds up better than Opus 5 against prompt-injection attempts. Business customers that Anthropic quoted echoed that same read from daily use rather than benchmark charts: GitHub said the model needed among the fewest tokens and steps it had measured across GitHub Copilot CLI and VS Code, while Stripe described a single Opus 5.5 session directing a dozen others through a stacked, 40-pull-request rebase, with every change clearing continuous-integration checks by the next afternoon.
The rollout also comes with fresh guardrails. Anthropic said Opus 5.5's biology and cybersecurity abilities sit close to those of Claude Mythos 5.1, so it is holding the new model to safeguards similar to what Claude Fable 5.1 already carries: vetted organizations can request access to Opus 5.5 now through Anthropic's Life Sciences Verification Program, and the company said a matching Cyber Verification Program for vetted cybersecurity practitioners will widen access over the coming weeks. Anthropic pitched Opus 5.5 as its first release since it publicly called on the industry to slow the pace of frontier development, adding that outside reviewers, including Frontier Design and METR, examined the model ahead of launch.
Claude Opus 5.5 is already reachable through Anthropic's own Claude Platform, listed under the model identifier claude-opus-5-5, and via the three big cloud marketplaces, Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic indicated that lower-cost Sonnet and Haiku variants sharing the same 5.5 generation will arrive within weeks, extending the same efficiency and safety work to its cheaper tiers.
