Anthropic's Claude Sonnet 5.5 rivals Opus 5.5, costs up to 30 percent less

Anthropic's Claude 5.5 family now has three named members: Fable, Opus, and the newly released Sonnet, with Haiku announced for the coming weeks. The company presents these as counterparts to OpenAI's GPT-6 Astra, Sol, and Luna, which kicked off the latest pricing battle. Roughly speaking, Opus sits a bit above Sol, Sonnet above Luna, and Fable above Astra, though Anthropic charges more across the board. Performance differences between matched tiers are small enough that cost may end up being the deciding factor.

The Sonnet 5.5 release leans heavily on that cost argument. On FrontierCode 1.1 at the "High" setting, Anthropic says the model scores ten points above Sonnet 5 at roughly one-fifteenth the cost per task. At low or medium effort settings, it already beats Sonnet 5's best scores on several benchmarks at about one-tenth the per-task cost. The catch is that finding the right effort level for each task is still more art than science, and the highest "Max" setting actually scores worse on FrontierCode than "Xhigh" because the model more often triggers a code-review function that splits work across sub-agents, sometimes causing timeouts or out-of-scope changes.

Anthropic is also adding safeguards to a Sonnet model for the first time because of the model's cybersecurity capability, and it has added classifiers against distillation attacks. Whether those measures work should become clear over the coming months. If Chinese labs have been benefiting from such attacks, closing that vector could widen the gap with Western labs again. The near-term question is whether independent testing confirms the speed and cost claims, and whether Haiku 5.5 arrives soon enough to help Anthropic close the price gap with OpenAI.