OpenAI and Anthropic have significantly reduced prices for their mid-tier AI models throughout 2026 to compete with open-source alternatives. On July 30, OpenAI lowered the cost of GPT-5.6 Luna by 80%, setting prices at $0.20 per million input tokens and $1.20 per million output tokens. The GPT-5.6 Terra model saw a 20% price reduction to $2 and $12 per million tokens.
Anthropic established permanent pricing for Claude Sonnet 5 on August 10 at $2 and $10 per million tokens. On September 22, OpenAI introduced GPT-6 Sol and Luna at $2/$10 and $0.10/$0.50, representing a 50% decrease compared to previous generations. Anthropic responded with Claude Opus 5.5, which offers users approximately 40% in effective savings through efficiency improvements.
Chinese AI firms continue to exert competitive pressure on pricing. DeepSeek V4 Flash is currently priced at approximately $0.14 per million input tokens and $0.28 per million output tokens. These models reportedly approach the performance of US mid-tier offerings while costing five to nine times less. Enterprises are responding by tiering workloads between cheaper routine models and premium specialized ones.
The aggressive pricing strategy aims to boost adoption and usage metrics ahead of anticipated IPOs for both OpenAI and Anthropic. While these cuts compress profit margins per API call, the companies are betting that increased volume will offset the losses. Despite these shifts, open-source models maintain a significant share of the market, particularly on platforms like OpenRouter.
