Anthropic Cuts Opus Costs 40%, OpenAI Answers in an Hour

Two of the largest AI labs cut the price of their flagship APIs on the same afternoon. On Sept. 22, 2026, Anthropic launched Claude Opus 5.5, and roughly an hour later OpenAI released GPT-6 Sol and GPT-6 Luna, according to 9to5Google's Ben Schoon, who published his account at 1:20 p.m. Pacific that day.

Anthropic's own pricing table puts Opus 5.5 at $4 per million input tokens and $20 per million output tokens, against $5 and $25 for Opus 5 — a 20% per-token cut. The bigger drop is on cached input: cache reads fall to $0.20 per million from $0.50, a 60% reduction, and cache writes to $5 from $6.25. Anthropic notes that cache reads "make up the majority of agentic and coding work costs." Stacking the lower per-token rate on top of fewer tokens used per task, the company says the model "will cost 40% less than Opus 5 on typical workloads" at default settings, while generating output more than 30% faster. A separate Fast mode for Opus 5.5 runs at $8 per million input and $40 per million output tokens.

OpenAI framed its release the same way. In a statement quoted by 9to5Google, the company said improvements "in caching and inference let us serve these models at lower cost, and we're passing those savings directly on to users and customers by reducing API prices for Sol and Luna by 50% compared with their GPT-5.6 promotional pricing." Neither Sol nor Luna is OpenAI's most capable model — that remains GPT-6 Astra. Luna prioritizes cost, while Sol sits between Luna and Astra on capability.

OpenAI's published statement gives the percentage rather than dollar figures. A pricing comparison posted the same day by developer Simon Willison lists GPT-6 Sol at $2 per million input and $10 per million output tokens, and GPT-6 Luna at $0.10 and $0.50 — figures that come from that secondary tally, not from OpenAI's own wording.

Why this matters beyond the two announcements: per-token rates are the direct input cost for everyone building on these APIs, and both labs moved in the same direction on the same day. For heavy coding and agent workloads, Anthropic's cache-read cut is the line item that moves budgets most.

Neither company said the timing was coordinated or that either release was a response to the other; what is documented is only that OpenAI's followed about an hour later. Some things are still open: Anthropic says Claude Sonnet 5.5 and Haiku 5.5 arrive "in the coming weeks" without pricing, and per 9to5Google, Sol and Luna were available in ChatGPT Work and Codex at launch but not yet in Chat.

Disclosure: NewUJ's editorial process uses Anthropic's Claude models.