Anthropic launches Claude Opus 5.5, outperforms Fable 5.1 at lower cost

Anthropic unveiled Claude Opus 5.5 on Tuesday, saying the new model outperforms its own Fable 5.1 on a suite of agentic benchmarks while costing roughly 60 percent less per API call. The company positioned Opus 5.5 as a “frontier” model built for long‑running coding agents, research tasks, and professional knowledge work.

The launch comes as developers and enterprises scramble for AI tools that can handle complex, multi‑step workflows without blowing up budgets. Anthropic’s earlier flagship, Mythos 5.1, was praised for its reasoning abilities but struggled with sustained code generation. Fable 5.1, released only weeks earlier, improved on that front but remained pricey for large‑scale deployments. Industry analysts say the pressure to lower inference costs has intensified after OpenAI rolled out GPT‑4 Turbo at a reduced price, prompting rivals to chase similar economics.

In internal tests, Opus 5.5 achieved a 12‑point lead over Fable 5.1 on AgentBench, a benchmark that measures an AI’s ability to plan, execute, and adapt in dynamic environments. On coding tasks, the model reduced error rates by 18 percent and completed scripts 22 percent faster. Anthropic reports that the API now runs at $0.002 per 1,000 tokens, compared with $0.005 for Fable 5.1. “We wanted a model that could stay on for hours, write reliable code, and still be affordable for businesses,” said Dario Amodei, Anthropic’s CEO, in a brief statement. Analysts at Loup Ventures called the price cut “significant enough to make Opus 5.5 the default choice for many SaaS providers.”

Anthropic says Opus 5.5 will be available to existing customers today and to new users via a public beta next week. The company plans to integrate the model into its Claude‑AI platform, allowing developers to plug it into IDEs, data pipelines, and research notebooks. If the early performance claims hold up, Opus 5.5 could shift the balance in the competitive race for affordable, high‑capability AI agents, pressuring rivals to match both speed and cost. The next few months will reveal whether the model’s promise translates into real‑world productivity gains for enterprises.