Anthropic has unveiled Claude Opus 5.5, the latest upgrade in its Opus model line, arriving roughly two months after the launch of Opus 5. The refreshed model brings notable improvements across performance, efficiency, and pricing. It’s now available in the Claude suite, with Opus 5.5 leading the pack while Sonnet 5.5 and Haiku 5.5 are set to follow.
What’s New in Opus 5.5
The Opus series is Anthropic’s top-tier model line, intended for demanding workloads. It sits above Haiku (small and budget-friendly) and Sonnet (mid-tier), and just under the even more powerful Fable/Mythos line. Opus 5.5 succeeds Opus 5 as the flagship Opus model.
Externally validated by evaluators like METR and Frontier Design, Opus 5.5 shines on Anthropic’s most rigorous alignment tests, achieving the company’s strongest scores yet. In benchmarks simulating large-scale tasks, users reported dramatic gains: a codebase migration involving 680,000 lines was completed in less than a day—something that previously would have taken weeks. When measured against Opus 5, Opus 5.5 fixed performance flaws in 39 out of 40 cases in testing that aimed to cut load times across every page of a web app, while Opus 5 made only modest gains and introduced changes to behavior. Graphic design tasks also saw Opus 5.5 producing higher polish than earlier models.
On long-form engineering work, Opus 5.5 delivers much faster results: auditing and repairing a 200,000-line codebase took under 3 hours, compared to over 20 hours with Opus 5, and required far fewer tokens. In translation and rewrite tasks, including translating HAProxy from C to Rust, Opus 5.5 beat even Fable 5.1—finishing in 9.5 hours vs. 12 hours and costing nearly half as much. On multiple benchmark suites—agentic coding, Terminal Bench 4.0, CursorBench—Opus 5.5 either matches or surpasses competitive models like GPT-6 Astra and GPT-5.6 Sol, all while operating at a fraction of their cost.
Efficiency and Access Upgrades
One of the biggest shifts is in cost. Opus 5.5 requires less computation to serve, translating into pricing reductions: input and output tokens are priced at $4 and $20 per million, respectively—about 20% less than Opus 5. Cache reads, a major cost driver in coding and agent tasks, drop by roughly 60%. In addition, output generation speed improves by over 30% relative to Opus 5. To sweeten the deal further, usage limits for Pro, Max, and Team plans have been raised to five hours, and subscription users now have a rate limit reset feature that they can deploy on their own schedule.
While Opus 5.5 sets the standard, Anthropic has confirmed that Sonnet 5.5 and Haiku 5.5 are coming soon. These models will offer many of the same gains in performance, safety, and efficiency as Opus 5.5.
Why this matters: Claude Opus 5.5 represents a significant step forward in balancing raw capability with cost-effectiveness. By delivering Fable-level performance in many areas while dropping the price and computational load, it broadens access to high-end AI workloads.
