OpenAI and Anthropic cut LLM prices by up to 80% amid open-weight competition and compute efficiency gains

OpenAI and Anthropic reduced LLM prices significantly between June and September 2026, with OpenAI cutting GPT-5.6 Luna prices by 80% and Anthropic reducing total costs for Opus 5.5 by 40-60%. Vercel data shows token costs falling to 43.3% of April 2026 levels.
Price cuts were driven by the commoditization of LLMs, low switching costs via tools like OpenRouter, and competitive pressure from open-weight models such as DeepSeek V4.1 Flash and Moonshot AI's Kimi K3.1 Pro.
Anthropic's Claude Opus 5.5 and OpenAI's GPT-6.1 Sol achieved significant cost reductions through improved compute efficiency, with cache read costs accounting for 96-99% of total token spend in agentic coding.
The pricing strategy coincides with rumored IPOs for both companies, creating a contradiction between the need for profitability and aggressive price cuts, though inference margins are reported to be as high as 80%.

Market Dynamics

Open-weight models like DeepSeek V4.1 Flash (Artificial Analysis index 45) and Kimi K3.1 Pro (index 55) provide economic checks on proprietary pricing, with Kimi K3.1 Pro offering comparable performance to Opus 5.5 at half the input/output cost.
The rapid obsolescence of high-end models like Fable 5.1 and GPT-6 Astra following the release of Opus 5.5 suggests a shift toward faster model iteration cycles.

OpenAI cut the usage allowance for its $100/month Codex tier by 50% alongside the GPT-6.1 Sol release, citing model efficiency gains.
Improved compute efficiency may reduce the demand for new data center construction, though inference remains only a portion of total compute requirements.