Anthropic Opus 5.5, OpenAI GPT-6 Sol and Luna Slash AI Costs

Summary

On September 22, 2026, Anthropic launched Opus 5.5, its flagship workhorse for coding and complex knowledge work. Input and output cost $4 and $20 per million tokens, 20 percent below Opus 5, while cache reads fall 60 percent to $0.20 and output arrives more than 30 percent faster. Anthropic claims typical savings near 40 percent because the model also uses fewer tokens. Anthropic and partner benchmarks show modest advantages over GPT-6 Astra in some coding and knowledge tasks, although OpenAI’s model, released earlier in September, had sometimes beaten Opus 5. Requests involving high-risk cybersecurity or biology content may be automatically and transparently routed to an older model under protections also used for Fable 5.1.

OpenAI released GPT-6 Sol and Luna, trained with methods similar to GPT-6 Astra. They improve some benchmarks by a few percentage points while costing half as much as their predecessors. Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50. OpenAI positions Astra as its powerful premium model, Sol as an efficient daily driver, Terra as balanced and general purpose, and Luna as fastest and cheapest. The launches target enterprises increasingly using model routers and open-weight alternatives to reduce frontier-model spending. Customers now prioritize predictable costs and well-orchestrated deployments, since useful AI still requires sophisticated contextualization, software harnesses and organizational practices.