OpenAI and Anthropic have lowered the costs of their primary artificial intelligence models by as much as 80 percent. These price reductions commenced in June, with some API usage costs falling by 75 percent or more.
By 30 July, the company reduced the prices for GPT-5.6 Luna by 80 percent to $0.20 for input and $1.20 for output. Anthropic made a temporary launch discount for Claude Sonnet 5 permanent in August.
In September, Anthropic cut Claude Opus 5.5 pricing by 40 percent to 60 percent. The company further reduced cached input pricing to 5 percent of the regular input cost.
Max Woolf said the speed of the reductions surprised him. Cache reads account for 96 percent to 99 percent of token spending in AI coding tools workflows.
Vercel’s AI model gateway reported that the average cost per token fell to 43.3 percent of its April 2026 level. The Flash V4.1 1017 model is available through 29 inference providers on OpenRouter.
Moonshot AI plans to release Kimi K3.1 Pro weights on 4 November. Anthropic could file for an IPO within the next few months.
OpenAI reduced the usage allowance for its $200-per-month Codex tier by 50 percent. Woolf claimed that GPT-6.1 Sol’s pricing reduced his effective usage costs by approximately 75 percent.
