Seven days after both companies called publicly for the industry to slow down, OpenAI and Anthropic released cheaper models within hours of each other.
What arrived
OpenAI extended the GPT-6 family downward. GPT-6 Sol is listed at $2 per million input tokens and $10 output for short context, against $4 and $20 for GPT-5.6 Sol. GPT-6 Luna comes in at $0.10 and $0.50, against $0.20 and $1.20 for the 5.6 generation. GPT-6 Astra stays on top at $10 and $50.
Anthropic released Claude Opus 5.5 the same day at $4 and $20, down from $5 and $25 for Opus 5, and says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run. Cached reads on Opus 5.5 are priced at 5% of base input rather than the usual 10%.
Claude Sonnet 5.5 followed on September 28 with no change in price: still $2 and $10. Anthropic's claim there is about speed rather than rate, saying output runs 30% faster and total cost per task drops by up to 30% because the model uses fewer tokens and fewer tool calls.
The number that matters
GPT-6 Sol and Claude Sonnet 5.5 now sit at exactly the same list price, $2 in and $10 out. Two competing mid-tier models from the two leading labs, priced identically, is not a coincidence. It is what a price war looks like when neither side wants to be the cheaper option by accident.
The awkward part
On September 15 both companies confirmed talks about coordinating on safety, including the pace of frontier development. A week later they shipped competing models on the same afternoon. Nothing here contradicts the letter of those talks, which concern frontier capability rather than price. But it does show which clock the industry actually runs on.
What to do about it
If you are paying for a mid-tier model, re-check your bill. A workload that cost $20 per million output tokens on GPT-5.6 Sol costs $10 on GPT-6 Sol, and the cheapest tiers moved further still. The saving is automatic only if you change the model name in your code, so it is worth an hour of somebody's time this week.