What happened
Sol and Luna fill out OpenAI's GPT-6 family, which began earlier this month with the flagship GPT-6 Astra. The two new models are aimed at different jobs. Sol is the mid-tier workhorse, built for complex work such as coding, technical analysis and multi-step tasks. Luna is the budget option, tuned for high-volume, everyday work like summarizing documents, pulling information out of text and answering quick questions.
Both are available in the OpenAI API under the names gpt-6-sol and gpt-6-luna, and are rolling out gradually to ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users get Luna in the desktop app.
The key numbers
- Pricing: Sol costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for the previous GPT-5.6 Sol. Luna drops to $0.10 input and $0.50 output, from $0.20 and $1.20.
- Caching: OpenAI says improved prompt caching gives up to a 90% discount on cached input tokens, and the cache now survives changes to reasoning effort and tool settings.
- Coding: On the DeepSWE v1.1 software-engineering benchmark, Sol scored 68.8%, just behind Anthropic's Claude Fable 5 at 69.9%, which OpenAI says it reaches at a fraction of the cost. Luna scored 66.6%.
- Reliability: OpenAI claims Sol makes about half as many factual mistakes as its predecessor, approaching the reliability of the far more expensive Astra.
- Honesty in coding: According to OpenAI's alignment testing, as reported by The New Stack, Sol's rate of misleading claims about its own coding work fell to 1.3% from 10.4%.
One caution: these benchmark figures come from OpenAI itself. Independent testing, including direct comparisons with Anthropic's brand-new Opus 5.5, has not caught up yet.
Why it matters
The biggest cost of running AI in real products is not the flashy flagship model; it is the millions of routine calls made by mid-range and budget models every day. Halving those prices changes the math for startups, customer-support tools, coding assistants and automated "agent" workflows that chain many model calls together. It also puts pressure on rivals: when two leading labs ship price-cutting models within the same afternoon, the market expectation becomes clear. Frontier-level performance is quickly becoming a commodity, and cost and reliability are turning into the main battleground.
Why it matters for developers and Africa's tech scene
For developers in Kigali, Lagos, Nairobi and beyond, API costs are often priced in dollars but paid from local-currency revenue, so every price cut matters more. A model like Luna, at ten cents per million input tokens, makes it far more realistic to build affordable chatbots in local languages, document-processing tools for small businesses, or tutoring apps for students. Better caching also helps teams on tight budgets who send the same large instructions or documents again and again. The practical advice: re-test your current prompts on the new models and compare cost per task, not just price per token.
What's next
Expect independent benchmark results over the coming days, especially head-to-head tests between GPT-6 Sol and Claude Opus 5.5. Developers will also be watching whether the rollout to ChatGPT goes smoothly and whether Google and other rivals respond with price cuts of their own.
Also in tech today
- Anthropic released Claude Opus 5.5, promising lower prices and performance close to its top Fable models.
- Hacking group ShinyHunters claims it breached the FBI and stole data on agents and job applicants; the claim has not been confirmed.
- Qualcomm unveiled two new smartphone chips built around on-device AI.
Sources
- OpenAI: Introducing GPT-6 Sol and Luna
- TechCrunch: OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes
- The New Stack: OpenAI releases GPT-6 Sol and Luna and cuts token prices in half
- TechCrunch: Anthropic releases Opus 5.5
- TechCrunch: ShinyHunters claims FBI breach
- TechCrunch: Qualcomm launches two new smartphone chips
0 Comments