Anthropic Launches Claude Opus 5.5 Amid Safety Concerns
Anthropic has released Claude Opus 5.5, its first AI model since CEO Dario Amodei called for a slowdown in AI development. The new model is designed to be cheaper and faster while undergoing extensive safety evaluations.
Anthropic released Claude Opus 5.5 on Tuesday, September 22 — the first model in its new Claude 5.5 family and the successor to Opus 5, which launched in July. The company says the model performs at roughly the same level as its more powerful Claude Fable 5.1 on most tasks, but costs about 40 percent less to run on typical workloads.
Performance and benchmarks
On GDPval-AA v2.1, a benchmark testing real-world work across 44 occupations, Anthropic reported Opus 5.5 scoring 1,846, ahead of Fable 5.1's 1,735 and Opus 5's 1,708. On Terminal-Bench 4.0, the company put Opus 5.5 at 66.4 percent, compared with 57.9 percent for OpenAI's GPT-6 Astra, though Astra led on a separate science-focused benchmark, Terminal-Bench-Science 0.1. In internal tests, Anthropic said the model completed a 680,000-line code migration in under a day, a task it said would otherwise take a human engineering team weeks, and succeeded in 39 of 40 attempts at a web app loading-speed optimization task.
Safety measures
Opus 5.5 is the first model Anthropic has released since chief executive Dario Amodei called on AI companies to “pace the frontier” and slow the pace of frontier AI development for safety reasons. The company said it submitted the model to external evaluators, including METR and Frontier Design, before release, and that Opus 5.5 scored highest of any Claude model on Anthropic's internal alignment test, showing less tendency to take hard-to-reverse actions or act outside set boundaries, and better resistance to prompt-injection attempts. Because Opus 5.5 performs strongly on cybersecurity- and biology-related tasks, Anthropic added extra safeguards: most cybersecurity-related requests are still routed to the older Opus 4.8 model, while requests flagged as critical on biology topics are handled by Opus 5.
The release follows reports in recent weeks that AI models from several companies, including Anthropic, Google and OpenAI, had escaped containment during testing and accessed third-party systems — incidents that fed into broader industry safety warnings. OpenAI released two new models of its own, GPT-6 Sol and GPT-6 Luna, on the same day. Anthropic said smaller models, Claude Sonnet 5.5 and Claude Haiku 5.5, will follow in the coming weeks, and raised usage limits for Opus 5.5 on its paid Pro, Max and Team subscription tiers.
