Anthropic ships Claude Opus 5.5 at 40% less to run than Opus 5

Anthropic framed the Opus 5.5 launch around capability at a lower price, saying it performs at Claude Fable 5.1's level for most tasks and costs 40% less to run than Opus 5. The company also credited efficiency work for the price cuts, and third-party signals such as Cline's report that Opus 5.5 beats Fable 5.1 on the Artificial Analysis Intelligence Index at about 2.5x lower cost point in the same direction. The launch was not just a model swap: Anthropic raised 5-hour session limits 20%, said lower pricing means limits go 25% further, and gave Pro, Max, and Team users a banked rate-limit reset they can use whenever they choose.

The pricing story is more complicated than the headline suggests. Vals notes Opus 5.5 often uses more tokens, especially on coding, and Artificial Analysis found that at max effort the model costs $5.98 per Intelligence Index task versus $5.86 for Opus 5. Higher token usage alone would have raised cost per task about 80%, to $10.51, before the base-price cut and cheaper cache reads brought it back down to $5.98. That is why the 40% cheaper claim applies to default (medium) settings, not max effort.

OpenAI's response came within hours, with GPT-6 Sol and Luna described as faster, cheaper models inheriting much of GPT-6 Astra's advances, priced at $2/$10 and $0.10/$0.50 per million input/output tokens. Commentators split on the day's winner: some saw Anthropic ahead on capability surprise, others said OpenAI won on price and deployment ergonomics. The outcome may hinge on whether buyers weigh headline model quality over per-task economics, and on whether the promised Sonnet 5.5 and Haiku 5.5 releases in the coming weeks shift that balance.