Anthropic and OpenAI Unveil Cheaper Claude Opus 5.5 and GPT-6 Models, Sparking AI Price War

Anthropic and OpenAI shipped cheaper frontier models within 90 minutes of each other, and the pricing is the news rather than the capability. Claude Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens, 20% below Opus 5 on a per-token basis; Anthropic frames the practical saving as roughly 40% on a typical workload because the model also consumes fewer tokens to reach an answer. Cache reads fell furthest, from $0.50 to $0.20 per million tokens, a 60% cut. That 60% figure is a cache-read number, not a blanket discount on the model.

OpenAI's response landed the same day. GPT-6 Sol and GPT-6 Luna are priced at half the cost of their GPT-5.6 equivalents, and OpenAI says Sol makes about half as many mistakes as its predecessor. Two vendors cutting frontier pricing on the same morning is the clearest signal yet that inference cost, not raw capability, is where the competitive pressure now sits.

The discounting is spreading outward from the model providers. Enterprise software vendors including Adobe, Amazon, Figma, HubSpot, Microsoft and Workday have introduced their own AI discount programs, aimed at customers who might otherwise route workloads directly to the cheaper Anthropic or OpenAI endpoints . The competitive read-through runs to the hyperscalers that host this inference, since cheaper tokens mean more volume at thinner unit economics. MSFT, AMZN, GOOGL and ADBE all sit on one side or other of that trade.

Capital is still flowing to the talent behind these models. Mirendil, founded by former Anthropic researchers, is in talks to raise up to $1 billion at a $5 billion valuation, with Kleiner Perkins leading and Andreessen Horowitz also in discussions. What to watch from here is whether the price cuts translate into usage growth large enough to offset the lower per-token take, and whether a second discount cycle follows. Note that the 'pacing the frontier' language around these releases traces to Anthropic's own stated caution about deployment speed, not to any regulatory action.