OpenAI and Anthropic Cut AI Model Prices as Competition Intensifies
OpenAI reduced GPT-5.6 Luna pricing by 80%, while Anthropic launched Claude Opus 5.5 at prices 20% below Opus 5.
OpenAI and Anthropic have cut prices for several artificial-intelligence models, intensifying competition over token prices, performance and operating costs as developers compare systems for specific workloads.
OpenAI reduced GPT-5.6 Luna pricing by 80% on July 30, to $0.20 per 1 million input tokens and $1.20 per 1 million output tokens. GPT-5.6 Terra pricing fell 20%, to $2 per 1 million input tokens and $12 per 1 million output tokens.
GPT-5.6 Luna is positioned for fast, high-volume workloads, while GPT-5.6 Terra is designed to balance capability and cost. The cuts affect different model tiers, with developers weighing price alongside output quality, token use and infrastructure requirements.
GPT-6 Sol launched Sept. 22 at $2 per 1 million input tokens and $10 per 1 million output tokens. GPT-6 Luna launched at $0.10 per 1 million input tokens and $0.50 per 1 million output tokens.
Claude Opus 5.5 launched Sept. 22 at $4 per 1 million input tokens and $20 per 1 million output tokens, versus $5 and $25 for Opus 5. Opus 5.5 is estimated to cost 40% less to run for typical token-billed workloads, a running-cost comparison distinct from its list-price reduction.
Claude Sonnet 5’s introductory pricing became permanent Aug. 10 at $2 per 1 million input tokens and $10 per 1 million output tokens.
DeepSeek-V4-Flash is priced at $0.14 per 1 million input tokens for cache misses and $0.28 per 1 million output tokens, adding another low-price option outside the two U.S. companies.
List prices do not capture the full cost of operating an AI system. Developers also account for how much work a model completes, the number of tokens required, caching, infrastructure, engineering and support expenses. The available announcements do not establish that developers are abandoning open-source or open-weight models because of the cuts.
OpenAI and Anthropic use internal or selected evaluations for performance comparisons. Those comparisons do not by themselves establish results across every task or workload. Competition should be assessed by the cost of completing specific tasks, not only by headline token prices.
