OpenAI launched GPT-6 on 22 September in two production tiers — Sol, its standard long-context reasoning model, and Luna, designed for high-frequency agentic workflows — while Anthropic published Claude Opus 5.5 within hours. It was the first time the two leading frontier AI labs have released competing flagship models on the same calendar day.
GPT-6 Sol is priced at $2 per million input tokens and $10 per million output — approximately 50 per cent below the launch pricing of GPT-5. Luna is priced at $0.10 input and $0.50 output, targeting enterprise agentic deployments where per-task cost governs adoption. OpenAI's published figures for Sol report a deception rate of 1.3 per cent on its standard alignment evaluation suite, compared with 10.4 per cent for its predecessor — the largest single-generation alignment improvement the company has reported.
Claude Opus 5.5 answered with a 40 per cent cost reduction against Opus 5, a Terminal-Bench score of 66.4 per cent (against Fable 5.1's 55.8 per cent on the same evaluation), and a one-million-token context window. Anthropic positioned it as the leading model for autonomous long-context work and multi-step agentic tasks with extended reasoning requirements.
The pricing convergence reflects the compute cost trajectory both labs have signalled for several quarters. Sol's 50 per cent price reduction — combined with the alignment improvement implied by its deception rate — reframes the frontier pricing conversation: the question is no longer whether frontier-quality reasoning is affordable at enterprise scale, but whether organisations have the deployment infrastructure to use it at the volume the economics now support.
GPT-6 generated the largest Google Trends spike in AI model search traffic for the week of 22–25 September.