OpenAI and Anthropic Are Giving Away Tokens Almost for Free

Over the past few months of 2026, OpenAI and Anthropic have cut prices on their top models. These are the two companies behind the leading AI chatbots and coding models. In July, OpenAI cut the price of GPT-5.6 Luna, its cheaper model, by 80%. In September, the company released GPT-6 Sol at half the price of the previous version. Anthropic cut the price of Opus 5.5, its best coding model, by 40-60%. For Fable 5.1, another leading Anthropic model, the cost of reading cached text dropped by 75%. Cached text is data reused from an earlier part of a conversation instead of being reprocessed each time. According to Vercel, the company that runs a gateway routing requests to different AI models, the average price per token among top models has fallen to 43% of its April 2026 level.

Lower prices mean every run of a coding agent in tools like Claude Code, a terminal-based AI coding assistant, now costs less for the same work. Part of this pressure comes from open-weight models. These are models whose base files are published, so anyone can run them. DeepSeek, a Chinese AI lab known for very cheap models, is a good example. These models do well on benchmarks. But they still fall short of Opus or Fable in real agentic coding work. So the price cuts look more like a result of cheaper infrastructure than pure DeepSeek-style competition. There is a timing detail worth noting. Both companies are preparing to go public: OpenAI in 2027, and Anthropic within a few months. Cutting their main revenue source right before investors start poring over their books makes you wonder how much per-token margin they are actually giving up.