OpenAI announced the release of two new GPT‑6 models—Sol and Luna—positioned as more affordable and reliable alternatives to its earlier GPT‑5.6 series. The rollout follows the earlier launch of GPT‑6 Astra, the lab’s flagship model, and comes amid a heated rivalry with Anthropic, which introduced a new Opus 5.5 version just before OpenAI’s announcement.
What happened
OpenAI’s blog post and a TechCrunch report detail the launch of GPT‑6 Sol and GPT‑6 Luna. Sol is marketed for complex, high‑skill tasks such as coding, while Luna targets high‑volume, goal‑oriented work like document summarization and quick question answering. Both models benefit from the same training methods used for GPT‑6 Astra, but they are optimized for speed and cost efficiency.
Key points from the announcement include:
- Pricing cut: API prices for Sol and Luna are reduced by 50% compared with the GPT‑5.6 promotional rates. Sol’s input cost drops from $4 to $2 per million tokens, and output from $20 to $10. Luna’s input falls from $0.20 to $0.10, and output from $1.20 to $0.50.
- Error reduction: Internal factuality tests show GPT‑6 Sol makes roughly half the mistakes of its predecessor, approaching Astra‑level reliability at a lower price point. Luna also shows substantial factuality gains.
- Coding performance: OpenAI cites a lower error rate for coding tasks, though exact figures are not disclosed.
- Availability: The models are live in ChatGPT Work, Codex for paid accounts, and the ChatGPT API. Luna is also accessible in the desktop app and to Free and Go users. OpenAI plans a gradual rollout to the ChatGPT web and app interfaces throughout the day.
- Competitive claim: OpenAI asserts that Sol and Luna outperform Anthropic’s top models—Fable and Opus—on a range of benchmarks.
Why it matters
The launch signals OpenAI’s push to democratize high‑performance AI by moving powerful capabilities down the cost curve. By halving API fees, OpenAI makes sophisticated language models more attainable for startups, developers, and enterprises that previously faced budget constraints. The reduction in factual and coding errors addresses longstanding concerns about AI reliability, especially for professional workflows where incorrect output can be costly.
Lower pricing also reshapes the economics of AI‑driven automation. OpenAI’s internal benchmarks, such as the AutomationBench test, show GPT‑6 Sol delivering higher scores than Anthropic’s Claude Opus 5 while costing only about 9% of Opus 5 per task. Similar gains are reported for Luna, which improves on its predecessor by over five percentage points at a 58% lower cost.
These improvements could accelerate adoption of AI agents in business processes, from sales and marketing to finance and HR, by reducing the financial barrier to large‑scale token consumption.
The bigger picture
OpenAI’s strategy of expanding the GPT‑6 family mirrors a broader industry trend: offering tiered models that balance capability, cost, and accessibility. The company’s emphasis on “cost‑intelligence curve” positioning reflects an ongoing competition with rivals like Anthropic, which released Opus 5.5 shortly before OpenAI’s announcement. Both firms are racing to demonstrate superior performance at lower price points, a dynamic that drives rapid iteration across the sector.
The launch also aligns with OpenAI’s public commitment to third‑party safety assessments. In a separate policy brief released the same day, OpenAI outlined priorities for independent evaluations of safety cases, safeguards, and risk mitigation. By providing deeper access to training data, model reasoning, and internal deployment practices, OpenAI aims to bolster confidence in its models’ alignment and security—an effort that becomes more critical as models become cheaper and more widely deployed.
What happens next
OpenAI plans to roll out Sol and Luna gradually across its ChatGPT platforms over the course of the day, making the models immediately available to most paid users via the API and to free users for Luna. The company also indicates that the models will be integrated into its broader suite of products, including Codex and ChatGPT Work, enabling developers to start building or migrating applications at the reduced cost.
Beyond the rollout, OpenAI’s statements suggest continued focus on improving factuality, coding reliability, and cost efficiency across the GPT‑6 line. The firm also reiterates its commitment to supporting independent third‑party assessments, which could shape future safety and governance practices for these models.
As the AI landscape evolves, the real impact of Sol and Luna will depend on how developers, enterprises, and researchers adopt the lower‑priced, higher‑reliability models—and whether competing labs can match or surpass these benchmarks in the months ahead.
