OpenAI Launches GPT‑6 Sol and Luna, Cutting API Prices by Half

OpenAI has expanded its GPT‑6 family with two new models, GPT‑6 Sol and GPT‑6 Luna. The company describes them as faster and more affordable versions of GPT‑6 Astra, the flagship it introduced earlier this month, trained with similar methods and carrying Astra's advances in professional work, factuality, coding, computer use and alignment to everyday use.

What OpenAI announced

According to OpenAI, work happens "at different scales, rhythms and budgets". Astra remains the most capable model and the choice for the most demanding projects, while Sol is aimed at heavier professional workloads and Luna at lighter tasks. The company says both models lead across the cost–intelligence curve, helped by improvements in caching and inference that let OpenAI serve them at lower cost.

API prices cut by half

Those savings are passed on directly. Prices per million tokens fall 50% against the models' GPT‑5.6 promotional rates: GPT‑6 Sol goes from $4 to $2 for input and from $20 to $10 for output, while GPT‑6 Luna drops from $0.20 to $0.10 for input and from $1.20 to $0.50 for output. OpenAI says the reduction makes advanced AI practical for more everyday tasks at scale.

Benchmark results

On AutomationBench, which tests agents on end-to-end business workflows across 47 tools in sales, marketing, operations, support, finance and HR, GPT‑6 Sol at extra-high reasoning effort scores 33.2% at $0.27 per task. OpenAI says that beats Claude Opus 5 at maximum effort — 26.9% — at 9% of its cost per task, and also exceeds Claude Fable 5.1 at a far lower price. GPT‑6 Luna at high effort improves on its predecessor by 5.4 percentage points while costing 58% less per task.

Other evaluations follow the same pattern. On Agents' Last Exam, Sol at maximum effort scores 56.4%, above Claude Opus 5's highest result at 60% lower cost per task. On OpenAI's internal factuality evaluation, Sol makes about half as many mistakes as its predecessor, approaching Astra-level reliability, while Luna at higher effort matches GPT‑5.6 Sol at about a hundredth of the cost. On DeepSWE v1.1, Sol reaches 68.8%, within 1.1 points of Claude Fable 5's 69.9%, at roughly 80% lower cost per task; on OSWorld 2.0 it scores 60.5% in computer use.

Availability and caching

Both models are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and GPT‑6 Luna also reaches Free and Go users in the desktop app; neither is in Chat yet. In the API they appear as gpt-6-sol and gpt-6-luna, with a gradual rollout through the day. OpenAI also reports improved prompt caching with higher hit rates and a 90% discount on cached input-token reads, and says GitHub measured a reduction of more than 50% in prompt tokens requiring fresh processing across billions of requests.