Skip to main content

OpenAI's GPT-6 Sol and Luna: Faster, Cheaper, and Surprisingly Capable

OpenAI's New Duo: GPT-6 Sol and Luna

OpenAI has quietly expanded its GPT-6 family with two new members: GPT-6 Sol and GPT-6 Luna. Trained similarly to the flagship GPT-6 Astra, these models bring top-tier performance in coding, factual accuracy, and computer use—but at a fraction of the cost. While Astra remains the premium choice, Sol and Luna offer a compelling alternative for those watching their budgets.

Image

Price Cut in Half, Costs Less Than 10% of Competitors

The API price for Sol and Luna is 50% lower than the promotional price of GPT-5.6. But the real eye-opener is the cost per task. In the AutomationBench cross-application business process test, GPT-6 Sol outperformed Claude Opus 5 at xhigh intensity, yet its cost per task was only 9% of Opus 5's. Even Luna, at high intensity, improved 5.4% over its predecessor while slashing cost per task by 58%.

In the Last Exam test for agents, Sol scored 56.4% under max effort—higher than Opus 5's best—and cut cost per task by 60%. Factual reliability? Sol's error rate is about half that of its predecessor, nearing Astra-level trustworthiness at a lower price. Luna also made significant strides.

Image

Performance Approaching Flagship, Longer Tasks More Efficient

For programming, both models are optimized for AI Coding Agent workloads. In FrontierCode 1.1 Main, Sol showed marked improvement over GPT-5.6 Sol and matched Claude Fable 5.1 xhigh at a lower cost. In DeepSWE v1.1, Sol scored 68.8% under max effort—just 1.1 points shy of Fable 5's 69.9%—with cost per task about 80% lower. Luna scored 66.6% under max effort, close to Opus 5 and Fable 5 at medium effort, but at 93% and 96% lower cost respectively.

Computer operations? Astra still leads, but Sol and Luna get the job done for less. In OSWorld 2.0 offline, Sol hit 60.5% under xhigh effort, nearly matching Opus 5's 60.3% at medium effort, with cost per task about 80% lower. Luna's max effort even surpassed GPT-5.6 Sol at medium effort, at roughly one-tenth the cost.

OpenAI also carried over Astra's improved communication style: clearer technical conversations, fewer jargon-y phrases, and tighter compression of low-value details. Plus, prompt caching for GPT-6 now has a higher default hit rate, helping agents reuse context and respond faster. Value alignment has improved too, with fewer misleading statements in coding tasks.

Availability

Starting today, GPT-6 Sol and Luna are available in ChatGPT Work and Codex for all Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access Luna in the desktop app. In the API, they're listed as gpt-6-sol and gpt-6-luna, but they're not yet in Chat.

Key Points

  • GPT-6 Sol and Luna are new cost-effective models from OpenAI, trained similarly to flagship Astra.
  • API prices are halved compared to GPT-5.6, with some tasks costing just 9% of competitors' rates.
  • Performance is close to flagship models in coding, computer use, and factual reliability, often at a fraction of the cost.
  • Improved communication style and prompt caching make them efficient for agent workloads.
  • Available now for most ChatGPT users, with API access under gpt-6-sol and gpt-6-luna.