OpenAI Launches GPT-6 Sol and Luna Minutes After Anthropic Drops Claude Opus 5.5
OpenAI launched two lower-cost GPT-6 variants, Sol and Luna, minutes after rival Anthropic released Claude Opus 5.5. The company also cut API prices for those models by about 50% versus GPT-5.6 promotional rates and positioned Sol and Luna as faster, cheaper options beneath its top-tier GPT-6 Astra.
Why It Matters
The rapid timing of these releases highlights intensifying competition between OpenAI and Anthropic, with both firms racing to deliver higher-performance models at lower prices for business-scale use. Price and performance moves could affect enterprise adoption and cost structures for AI-powered workflows across industries.
Key Facts
- Release date: Tuesday (reported September 22, 2026) / minutes after Anthropic launched Claude Opus 5.5
- New models: GPT-6 Sol and GPT-6 Luna
- Relation to Astra: Sol and Luna sit below GPT-6 Astra as cheaper, faster options
- Price cuts vs GPT-5.6 promotional rates: API prices reduced by ~50% per token
- Sol pricing (input/output): $2 per million input tokens; $10 per million output tokens (down from $4 and $20)
OpenAI unveiled two new GPT-6 family members, Sol and Luna, and simultaneously lowered their per-token API prices to roughly half the promotional rates charged for GPT-5.6. The company described the pair as streamlined alternatives to its flagship GPT-6 Astra, aimed at everyday business tasks where speed and cost-efficiency matter more than top-tier capability. OpenAI timed the announcement within minutes of Anthropic’s rollout of Claude Opus 5.5, underscoring a tight competitive cadence between the two firms. Sol and Luna are intended to mirror a tiered product strategy similar to Anthropic’s lineup: Astra corresponds to Fable, Sol to Opus, Luna to Haiku, and so on, with each model balancing performance and price for different use cases. The new pricing structure makes Sol substantially cheaper than prior rates: input tokens are $2 per million and output tokens $10 per million, down from $4 and $20; Luna drops to $0.10 and $0.50 per million for input and output, down from $0.20 and $1.20. OpenAI said these cuts leave it cheaper on a per-tier basis than Anthropic. OpenAI also released benchmark comparisons it said favor Sol on several practical tests. On AutomationBench, which measures an agent’s ability to complete end-to-end business workflows across 47 tools, GPT-6 Sol posted a 33.2% pass rate at its highest reasoning setting at an estimated $0.27 per task; OpenAI reported that Claude Opus 5 scored 26.9% and cost over 11 times as much per task on the same test. Additional evaluations cited include Agents’ Last Exam and OSWorld 2.0, where OpenAI reported further cost and performance advantages for Sol in certain settings. Both GPT-6 Sol and Luna are available now in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu subscribers; Luna was additionally rolled out to free and Go desktop app users. OpenAI said it was gradually deploying the models across services to maintain stability and that neither new model had reached the basic ChatGPT mobile app at the time of the announcement.
Keep Reading
Meta deal will help plastics recycling startup MacroCycle build its first factory
Singapore’s Nexstrom wants to bring 2D semiconductors to chip fabs
4 days to save up to $200: Reason 2 of 5 to be at TechCrunch Disrupt 2026
