Models

OpenAI and Anthropic Launch Models in New Price War

Anthropic has launched Claude Opus 5.5 and OpenAI has released GPT-6 Sol and Luna, triggering a dramatic price war that slashes developer costs for high-performance AI models.

Simon Willison1 day agoModels
Image: Simon Willison

Following the recent releases of Grok 4.7 and MiMo v2.6 Flash/Pro, Anthropic kicked off the latest round of AI competition by releasing Claude Opus 5.5, followed quickly by OpenAI's launch of GPT-6 Sol and GPT-6 Luna. This double release has triggered a massive price war. OpenAI has priced GPT-6 Sol and Luna at half the cost of their GPT-5.6 equivalents. For instance, GPT-6 Luna costs just $0.10 per million input tokens and $0.50 per million output tokens, making it one of OpenAI's cheapest releases ever, trailing only GPT-4.1 Nano at $0.10/$0.40 from April 2025 and GPT-5 Nano at $0.05/$0.40 from August 2025. Meanwhile, GPT-5.6 Terra is now priced identically to GPT-6 Sol, and GPT-5.6 has a scheduled 25 percent price increase coming in November. Grok 4.7, which originally cost $2 per million input and $6 per million output, is now equal to GPT-6 Sol on input.

Anthropic responded by cutting prices for Claude Opus 5.5. While previous versions from Opus 4.5 through Opus 5 cost $5 per million input tokens and $25 per million output tokens, Opus 5.5 drops those rates by 20 percent to $4 and $20 respectively. Additionally, Anthropic slashed the cost of cache reads by 60 percent, a major benefit for multi-turn agentic workflows. Opus 5.5 is designed to address communication style complaints, offering the intelligence of Fable 5.1 with better token efficiency and improved performance in Blender. However, at its "max" thinking level, Opus 5.5 hit its 128,000 output token limit and failed to complete a standard SVG generation test, costing $2.56 and taking nearly 20 minutes per attempt.

For developers, these shifts radically alter the economics of building AI applications. GPT-6 Luna is already proving highly competent at generating SQL queries and HTML, offering a massive discount compared to older options like Haiku 4.5, which remains at $1 per million input and $5 per million output. At the high end, GPT-6 Astra and Claude Fable 5.1 remain at $10 per million input and $50 per million output, leaving the mid-tier as the primary battleground. With Anthropic planning to release Sonnet 5.5 and Haiku 5.5 soon, the pressure on pricing is expected to intensify even further.

This is our own summary of reporting by Simon Willison

More in Models