Anthropic 发布 Claude Opus 5.5 - 模型在关键编程测试中击败 GPT-6 Astra

⚡️ Anthropic has released Claude Opus 5.5 - the model beats the GPT-6 Astra in key programming tests. According to Anthropic's own benchmarks, the new Opus scored 66.4% in Terminal-Bench 4.0 versus 57.9% for the GPT-6 Astra. In FrontierCode 1.1 the gap is smaller - 54.4% versus 53.3%, and in CursorBench 4.0 the new product received 57.8%. At the same time, Anthropic made the flagship cheaper. The API now costs $4/$20 per million input and output tokens instead of the previous $5/$25. Reading from the cache fell even more in price - from $0.50 to $0.20. Regular subscribers also received a bonus: the five-hour limits on Pro and Max were increased, and they were also given one additional limit reset, which can be used at any time. Source