OpenAI expands GPT-6 family with Sol and Luna models at 50% lower API prices

OpenAI has expanded the GPT-6 model family with GPT-6 Sol and GPT-6 Luna, two new models aimed at making the capabilities introduced in GPT-6 Astra accessible at lower cost. Both are priced 50% below their GPT-5.6 equivalents, continuing what has become a sustained trend of API price compression across the major AI providers.

GPT-6 Sol is positioned as the mid-tier model for demanding professional work, priced at $2 per million input tokens and $10 per million output tokens — down from $4 and $20 for GPT-5.6 Sol. GPT-6 Luna, the lighter option, drops to $0.10 input and $0.50 output per million tokens, from $0.20 and $1.20.

According to OpenAI, both models were trained using similar methods to GPT-6 Astra, bringing Astra's advances in factuality, coding, computer use, and alignment to more affordable price points. On AutomationBench, GPT-6 Sol at extended effort outperforms Claude Opus 5 at maximum effort while costing roughly 11 times less per task. On Agents' Last Exam, a long-horizon professional workflow evaluation, GPT-6 Sol achieves 56.4% at 60% lower cost per task than Claude Opus 5.

Factuality improvements are meaningful: on OpenAI's internal evaluation of real user-flagged errors, GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol, approaching the reliability level of GPT-6 Astra. GPT-6 Luna, while positioned for lightweight tasks, matches GPT-5.6 Sol at higher effort levels at a fraction of the cost.

On coding, OpenAI cites exponential internal growth in AI-assisted development — noting that daily token usage has exceeded $600 for the median internal researcher and $7,000 for the 90th percentile. Sol and Luna's lower prices are directly intended to make sustained coding agent use more practical for developers and organizations.

The pricing moves are significant in the context of the current race. OpenAI's Sol matches or outperforms Claude Fable 5.1's workflow benchmark scores at dramatically lower API cost. The company explicitly highlights that Fable 5.1's published costs understate its true price because they omit Claude Opus 5 fallback calls, which occur on roughly 40% of tasks in the benchmark.

GPT-6 Astra remains OpenAI's top model. Sol and Luna join it as the second and third tiers in what is becoming a fully stratified GPT-6 product line, roughly mirroring Anthropic's Opus/Sonnet/Haiku structure.