Anthropic introduced Claude Opus 5.5, the first model in its new Claude 5.5 family, at $4 per million input tokens and $20 per million output tokens. That is 20% less than the $5 and $25 rates of its predecessor, Opus 5.
OpenAI countered with GPT-6 Sol at $2 input and $10 output per million tokens — half the price of the GPT-5.6 Sol it replaces — alongside a smaller everyday model, GPT-6 Luna, at $0.10 input and $0.50 output.
Cache pricing is the real battleground
Beneath the headline rates, the deeper cut was Anthropic's. Cache reads — the price developers pay when a model reuses text it has already processed, which is central to long-running AI agent workflows — fell 60%, from $0.50 to $0.20 per million tokens.
Anthropic says Opus 5.5 will run typical workloads for 40% less than Opus 5, while generating output more than 30% faster.
OpenAI's numbers tell a sharper story on paper. GPT-6 Sol's new $2 and $10 rate is not just half of its own predecessor's price — it is exactly half of Anthropic's discounted Opus 5.5 pricing, and matches what Anthropic charges for its cheaper Sonnet tier.
An OpenAI spokesperson confirmed the rates are permanent list prices rather than a launch promotion, attributing them to improvements in inference and caching efficiency that the company is passing on to customers.
The benchmark crown now comes with a discount
Notably, Opus 5.5 has moved straight to the top of the Artificial Analysis Intelligence Index with a score of 58, ahead of GPT-6 Astra and Claude Fable 5.1 at 53. It is a first for the premium tier of the market: the independently benchmarked leader is now cheaper than the model it replaced.
Anthropic also raised five-hour usage limits on its Pro, Max, Team and seat-based Enterprise plans, signalling the cut extends beyond API developers to subscription users.
Not everyone followed the downward trend. xAI moved in the opposite direction, doubling its rate for long prompts.
The split suggests two competing bets on the economics of frontier AI: cheaper intelligence to win volume and lock in developers, versus premium pricing for long-context work that few rivals can match.
What it means for builders
For developers running agents, coding tools and research loops through these APIs, the per-token bill just went down regardless of provider. But the structure of the cuts matters as much as their size.
Anthropic's aggressive cache pricing rewards workloads that reuse context across long sessions, while OpenAI's flat halving favours raw throughput. The tokens got cheaper — and the choice of provider just got more complicated.
