Key Takeaways
- OpenAI and Anthropic have reduced prices for their flagship AI models by up to 80%.
- The price cuts began in June and have significantly lowered API usage costs.
- Competitors like DeepSeek and Moonshot AI are offering cheaper alternatives.
OpenAI and Anthropic have significantly reduced the prices of their AI models by up to 80% over the past four months, as competition from cheaper open-weight models intensifies.
The latest round of reductions began in late June when OpenAI launched GPT-5.6 Luna at $1 per million input tokens and $6 per million output tokens, reducing both prices by 80% to $0.20 and $1.20 by July 30.
Anthropic followed suit, making its temporary launch discount for Claude Sonnet 5 permanent and further reducing Claude Opus 5.5 pricing by 40% to 60% in September.
In a bid to remain competitive, OpenAI introduced GPT-6.1 Sol with half the input and output prices of its predecessor and reduced cached input pricing to 5% of the regular input cost, matching Anthropic’s rate.
Max Woolf, a former senior data scientist at BuzzFeed who tracks AI model pricing, noted that the speed of the reductions surprised him, adding that the cost of using leading AI models has fallen much faster than expected a year ago.
The lower prices are particularly important for AI coding tools such as Codex and Claude Code, which repeatedly access cached conversation histories, with cache reads accounting for 96% to 99% of token spending in such workflows.
A 75% reduction in cache-read prices can reduce effective usage costs by approximately 70% to 75%, according to data from Vercel’s AI model gateway, which shows that the average cost per token has fallen to 43.3% of its April 2026 level.
Despite the price cuts, total spending continues to increase due to improvements in AI model capabilities driving greater usage.
OpenAI and Anthropic are also facing stronger competition from open-weight AI models, particularly those developed by DeepSeek and Moonshot AI.
DeepSeek’s Flash V4.1 1017 model, released in April, offers input, output, and cached token pricing of $0.15, $0.60, and $0.003 per million tokens, respectively, and is now available through 29 different inference providers on OpenRouter.
Moonshot AI’s Kimi K3.1 Pro reportedly delivers performance comparable to Claude Opus 5.5 at half the input and output price and plans to release the model’s weights on November 4.
