A look at OpenAI's and Anthropic's aggressive price cuts from June through September, driven by low switching costs and pricing pressure from open-weight models

The cost of using popular LLMs has dropped dramatically, making advanced AI capabilities more accessible for any project.

Between June and September 2026, OpenAI and Anthropic engaged in a series of significant price reductions for their large language models (LLMs). OpenAI launched GPT-5.6 models, including the affordable Luna at $1/6/0.10, and later cut Luna's price by 80% to $0.20/1.20/0.02. They also reduced the price of GPT-5.6 Sol by about 20%. Anthropic introduced Claude Sonnet 5 at $2/10/0.20 and later made this price permanent, while also releasing Claude Opus 5 at $5/25/0.50. In September, Anthropic launched Fable 5.1 with a 75% cache read cost reduction and Claude Opus 5.5 at a lower price. OpenAI responded with GPT-6 Astra and later GPT-6 Sol and GPT-6 Luna at drastically reduced prices, including a 75% cumulative cost reduction for the latter family by halving cache prices. These cuts reflect intense competition, particularly from lower-cost open-weight models, which have driven down effective token costs across the market.

The aggressive price cuts by OpenAI and Anthropic, particularly the 80% reduction in GPT-5.6 Luna's price, were primarily a response to the growing threat from increasingly competitive open-weight models. Services like OpenRouter host numerous providers for open-weight models, making it easy for developers to switch based on price and performance, thereby reducing switching costs and creating pricing pressure on proprietary LLMs. This market dynamic forces major players to continuously lower prices to remain competitive, even as they introduce new, more capable models.

The sustained price reductions signal a potential commoditization of foundational LLMs, compelling companies like OpenAI and Anthropic to focus on building defensible moats. Strategies such as developing enterprise-specific solutions, enhancing personal agentic assistants, and potentially integrating models more deeply into ecosystems are likely to become more critical. The ongoing competition from open-weight models, which are rapidly improving in performance and supported by a growing number of inference providers, suggests that price wars may continue, impacting profitability for all players.