Anthropic announced that Claude Opus 5.5 leads in agentic coding and knowledge work, and costs 40% less to run than Opus 5 on typical workloads.
Anthropic introduced Claude Opus 5.5 on September 22, 2026, calling it the company’s strongest Opus model for agentic coding and knowledge work. On Terminal-Bench 4.0 it scored 66.4%, versus 52.3% for Opus 5 and 57.9% for OpenAI’s GPT-6 Astra; on GDPval-AA v2.1, which tests work across 44 occupations, it scored 1,846 Elo, ahead of Opus 5’s 1,708. Anthropic says typical token-billed workloads cost 40% less than with Opus 5: API rates fell 20%, from $5 to $4 per million input tokens and $25 to $20 per million output tokens, while cache reads dropped from $0.50 to $0.20 per million. The reduction matters to developers running long-lived coding agents and businesses automating knowledge work, where repeated context and output can make inference expensive. Anthropic attributes the larger workload saving to lower prices and fewer tokens per task; it also reports output generation is over 30% faster.