Anthropic Launches Opus 5.5, Claiming Its Best Benchmark Scores Yet at Lower Prices

Anthropic released Opus 5.5, a new flagship model the company describes as its strongest-performing system to date, and paired the launch with lower prices than its predecessor. The move puts fresh pressure on rivals at the top of the frontier-model market, where capability gains have often come with rising inference costs. Anthropic has not yet published a full benchmark table, but the price cut is the clearest signal that competition is now being fought on cost as much as raw scores.

The company's own framing — that Opus 5.5 is the best model it has evaluated — is notable because Anthropic typically reserves that language for internal comparisons against both prior Claude versions and competing systems. If the claim holds up under independent testing, it would mark another step in the rapid cadence of flagship releases that has defined the past 18 months. The reference to 'Fable-level performance' in early coverage suggests the model is being positioned against a high bar set by a leading competitor, though Anthropic has not confirmed which system that comparison targets.

Pricing is where the announcement could have the widest practical effect. Lower per-token costs make it cheaper for developers to run long-context tasks, agentic workflows and high-volume production traffic, the workloads that burn through budgets fastest. Startups that had been routing routine jobs to smaller, cheaper models may now reconsider using a top-tier system for more of their pipeline. That shift, if it happens at scale, would reshape how AI products are priced and packaged.

Anthropic has been competing aggressively for enterprise and developer share against OpenAI, Google and Meta, and price cuts have become a standard lever in that fight. The company's Claude family has leaned on safety positioning and long-context handling as differentiators, while rivals have pushed multimodality and distribution. A cheaper flagship narrows one of the main reasons buyers split traffic between vendors.

The obvious next question is verification. Independent evaluators, from academic benchmark groups to developer communities, will want to reproduce the performance claims and check whether the lower price comes with hidden limits on rate, context or throughput. Anthropic has not said when detailed model cards or third-party evaluations will appear. Until those land, the launch is a strong claim backed by a price tag — and the market will decide how much of it to believe.