Anthropic upgrades Claude with new Opus 5.5 model, details here

Anthropic has introduced its latest AI model, Claude Opus 5.5, following a two-month gap since the previous Opus upgrade. This new version joins the Claude family of models, which includes Haiku (least expensive), Sonnet (medium size and cost), Opus (largest and priciest), and the even more capable but costly Fable/Mythos. The release of Opus 5.5 comes after Anthropic called for a controlled approach to frontier AI development, emphasizing the model's alignment with ethical standards.

According to Anthropic, Opus 5.5 marks a significant leap forward from its predecessor, with external evaluators, such as METR and Frontier Design, confirming its robust performance. In a test, Opus 5.5 completed a 680,000-line code migration in under a day, a task typically taking an engineering team weeks. The model excels in identifying and correcting software inefficiencies - cutting web app load times on 39 out of 40 occasions, while Opus 5 achieved minimal improvements.

When tasked with building a game from a single prompt, Opus 5.5 outshone other models in terms of graphical quality and polish.

Opus 5.5 particularly shines in lengthy coding projects such as codebase-wide migrations and audits. An early user successfully audited and fixed a 200,000-line codebase in under three hours, a feat Opus 5 took over 20 hours to accomplish at a 2.5x token usage rate. In an internal test, Opus 5.5 translated HAProxy, widely used software for load balancing web traffic, from C to Rust, completing the task in 9.5 hours compared to Fable 5.1's 12 hours, while also reducing costs by 51%.

The model's performance surpasses GPT-6 Astra, the latest from OpenAI, on FrontierCode at a fraction of the cost. On Terminal Bench 4.0, Opus 5.5 matches Astra at roughly 40% of the cost, while it beats GPT-5.6 Sol by 11 points on CursorBench at a third of the price. Additionally, Opus 5.5 is priced more affordably than Opus 5, costing 40% less for typical workloads due to its lower compute requirements. Moreover, the model generates output 30% faster than Opus 5.

Anthropic is also adjusting the pricing structure for Opus 5.5 compared to Opus 5, with input and output tokens now priced at $4 and $20 per million, respectively, representing a 20% reduction from Opus 5. Cache reads, which account for most agentic and coding work costs, are now priced at $0.20 per million tokens, a 60% decrease from Opus 5's rate.

In response to user demand, Anthropic is increasing five-hour usage limits on Pro, Max, and Team plans and offering subscription users a resetable rate limit. The upgraded Claude Opus 5.5 model is now available, and Anthropic expects to roll out upgraded Sonnet 5.5 and Haiku 5.5 models for Claude in the coming weeks, offering similar performance, efficiency, and safety enhancements.