OpenAI launches GPT-6 Sol and Luna, roughly halving GPT-5.6-era prices
2026-09-22
About an hour after Anthropic's Opus 5.5 launch, OpenAI released GPT-6 Sol and GPT-6 Luna, which it says are 'cut from the same cloth' as GPT-6 Astra. Per pricing reported by Simon Willison, GPT-6 Luna is priced at $0.10/$0.50 per million input/output tokens (half of GPT-5.6 Luna's $0.20/$1.20), and GPT-6 Sol is $2/$10 (half of GPT-5.6 Sol's $4/$20). OpenAI also shipped improved prompt caching for GPT-6, including higher cache-hit rates, new diagnostics, and explicit breakpoints. TechCrunch reports OpenAI says the new models make 'fewer mistakes.'
Significance 4: A roughly 50% price cut across a model tier, timed directly against Anthropic's own cut, is a major pricing change that could move the cost axis and shift the bulk-cheap and hard-problems calculus, capped at 4 pending independent verification of OpenAI's 'fewer mistakes' claim.
Implications · machine-drafted, not owner judgment
GPT-6 Luna's $0.10/$0.50 pricing undercuts the closed-model price floor and puts pressure on DeepSeek V4's cost advantage in the bulk-cheap use case if quality holds up independently; more broadly, three labs cutting prices within days of each other (Anthropic, OpenAI, xAI) signals price competition is now structural rather than episodic, which compresses margins across the industry and lowers the cost floor for the owner's own agentic-tooling experiments.
- Artificial Analysis publishes independent pricing/speed/quality benchmarks for GPT-6 Sol and Luna
- DeepSeek or another open-weights lab responds with a further price cut
- GPT-6 Luna's real-world token usage doesn't erode the 50% headline cut the way Opus 5.5's did
Sources
- primary Introducing GPT-6 Sol and Luna retrieved 2026-09-23
- primary Better prompt caching for GPT-6 retrieved 2026-09-23
- press OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes retrieved 2026-09-23
- curator Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war retrieved 2026-09-23
Models
Affected axes
Hard reasoning · Cost · Output speed