OpenAI and Anthropic Just Cut AI Prices Up to 80%. Your Stack Is Probably Overpaying.
Since late June, OpenAI and Anthropic have repeatedly slashed token prices - up to 80% on some models. GPT-5.6 Luna dropped 80% in July. GPT-6 Sol and Luna were each halved on September 22. Claude Sonnet 5's launch discount became permanent in August. Fable 5.1 cut cache read costs by 75% on September 1, and because cache reads are 96-99% of total token spend in agentic coding workloads, that single change cut real-world cost by 70-75%. Opus 5.5 landed a 40-60% total reduction on September 22 while scoring better than Fable 5.1. The reason is structural: open-weight models like DeepSeek V4 Pro and Kimi K3 - hosted by more than 20 inference providers, with Kimi K3.1 Pro dropping November 4 - cap how much proprietary vendors can charge for a token. Vercel's AI gateway data shows cost per token down to 43.3% of April 2026 levels, and yet aggregate spend is still climbing because the models get better faster than they get cheaper. For any business running AI agents in production, the real question is no longer whether AI is worth it, but whether you are still paying June's prices for September's capability.
Published · By Alfred Belvedere
Today's trend: LLM API price war (Jun-Oct 2026): OpenAI and Anthropic cut flagship model prices up to 80% - GPT-5.6 Luna cut 80% on Jul 30 ($0.20/$1.20/$0.02), GPT-6 Sol/Luna halved on Sep 22 ($2/$10/$0.20, $0.10/$0.50/$0.01), GPT-6.1 Sol cut cache reads 50% on Oct 6 (75% cheaper cumulatively), Claude Sonnet 5 launch pricing made permanent Aug 10, Fable 5.1 cache read cost cut 75% Sep 1 (70-75% total cost drop since cache is 96-99% of agent token spend), Opus 5.5 total cost down 40-60% Sep 22. Root cause: open-weight models (DeepSeek V4 Pro, Kimi K3) cap pricing power - 20+ inference providers host Kimi K3 and Kimi K3.1 Pro drops Nov 4. Vercel AI gateway data: cost-per-token down to 43.3% of April 2026 levels, yet aggregate spend still rising because models improve faster than they get cheaper.