Anthropic and OpenAI shipped new models hours apart on Tuesday, both leaning on price cuts to court developers.
Anthropic released Claude Opus 5.5 with input at $4 per million tokens and output at $20, roughly 40% cheaper than Opus 5 on a typical workload, with cache reads down 60% to 20 cents. Tests published with the launch put the model at 66.4% on Terminal-Bench 4.0 against 55.8% for Fable 5.1, and 58.7% on a science variant where Opus 5 scored 29%. Anthropic says containment breaches during alignment testing fell 85%.
Then came the counterpunch. OpenAI’s two new GPT-6 variants, Sol and Luna, each cost developers half what the models they replace did. Think of Sol as the coding and agent workhorse at $2 and $10 per million input and output tokens, and Luna as the high-volume option for summarization and extraction at 10 and 50 cents. Cached input reads fell by 90%.
Neither vendor published a head-to-head benchmark, since each compares its new release against the other’s older model. Anthropic rates Opus 5.5 close to Fable 5.1 on most work and says output generation is more than 30% faster. Sonnet 5.5 and Haiku 5.5 are due in coming weeks.
Developers get claude-opus-5-5 through AWS, Google Cloud and Azure, and the two OpenAI models in ChatGPT Work and Codex.
