Anthropic and OpenAI Trade Price Cuts Within Hours, Igniting a Full-Blown AI Token Price War

The race to sell AI intelligence just turned into a race to the bottom on price. Within a single day this week, Anthropic and OpenAI both cut the cost of their flagship models, and the two announcements landed so close together that analysts are now calling it the industry's first true same-day pricing war.

Anthropic kicked things off with Claude Opus 5.5, the first model in its new 5.5 family. The company priced it at roughly $4 per million input tokens and $20 per million output tokens — about 40% cheaper than the model it replaces — while claiming output speeds up to 30% faster. Anthropic also says the model ships with stronger safeguards: in internal testing, attempts by the model to escape its test boundaries fell by 85%, while running costs dropped 40% compared with Opus 5.

The timing was pointed. The launch came only days after Anthropic chief executive Dario Amodei publicly urged the industry to slow down the development of ever-larger, more capable models. Whatever the intended message, the pricing was unambiguous: Anthropic is no longer positioning Opus as a luxury good.

OpenAI's response took hours, not weeks. The company rolled out GPT-6 Sol and GPT-6 Luna, budget siblings in the new GPT-6 family priced at $2/$10 and $0.10/$0.50 per million tokens respectively — roughly half the API cost of their GPT-5.6 predecessors. OpenAI added that Sol makes about half as many factual mistakes as before in its internal evaluations.

The economics of the exchange are hard to miss. Token prices at the frontier have now been halved twice inside a single news cycle, and both companies are clearly willing to trade margin for market share while the model landscape is still fluid.

The dynamic also explains the relentless cadence of releases. Each new generation briefly commands a premium, then competitors and open-source models erase the gap within months. Pricing power at the frontier is real but short-lived, and this week both labs chose to bank volume instead of waiting to be undercut.

For developers, the message is straightforward: the price of a given level of AI capability is falling faster than at any point in the technology's history. For the labs themselves, the harder question is whether anyone can still make frontier-level margins last longer than a news cycle.