OpenAI Introduces GPT-6 Sol and Luna: New Models Are Cheaper and More Powerful Than Their Predecessors

OpenAI has officially expanded the GPT-6 family with two new models — GPT-6 Sol and GPT-6 Luna. They leverage some of the technologies developed for the flagship GPT-6 Astra, but are designed for mainstream workloads with significantly lower operational costs.

The company positions GPT-6 Sol as a model for complex professional tasks, coding, and agentic workflows where Astra’s power is unnecessary. Luna sits at a more affordable tier and is primarily intended for low-cost, high-volume workloads.

Pricing is the headline change. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, down from $4 and $20 for GPT-5.6 Sol.

GPT-6 Luna is priced at $0.10 for input tokens and $0.50 for output tokens per million, down from $0.20 and $1.20, respectively. As a result, input costs for Luna dropped by exactly 50%, while output costs decreased by approximately 58%.

At the same time, OpenAI’s announcement characterizes the new pricing overall as a 50% cost reduction compared to GPT-5.6 generation promotional rates.

Sol Gains Precision and Significant Boost in Coding

According to OpenAI’s internal benchmark for factual accuracy, GPT-6 Sol makes roughly half as many errors as GPT-5.6 Sol, approaching the performance of the far more expensive Astra on this metric.

The company emphasizes that the test is based on anonymized ChatGPT conversations where users previously reported factual errors. Therefore, these results cannot be directly generalized to all query types.

A noticeable gain was also reported in coding performance. On DeepSWE 1.1, GPT-6 Sol scored 68.8% at maximum reasoning effort, approaching Claude Fable 5’s top result in OpenAI’s comparison. GPT-6 Luna scored 66.6% at its maximum effort level.

Sol and Luna also improved their performance-to-cost ratio on computer use benchmarks, though GPT-6 Astra remains OpenAI’s flagship model.

Cached Context Gets Up to 90% Price Cut

Along with the models, OpenAI updated its prompt caching system. Reused context can now receive up to a 90% discount on input token reads, which is particularly critical for long conversations and agents that regularly process the same large context base.

Sol and Luna also inherited Astra’s updated response style. OpenAI promises less jargon and awkward phrasing, clearer answers, and fewer low-value details without a noticeable loss of substance.

Models Not Yet Available in Standard Chat Mode

At launch, GPT-6 Sol and Luna are available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access GPT-6 Luna through the desktop app.

However, OpenAI specifically notes that the new models are not yet available in standard Chat mode. The company plans to roll out broader availability gradually.

For developers, both models are already available via the API under the model IDs gpt-6-sol and gpt-6-luna.

Ultimately, GPT-6 Sol and Luna do not replace Astra, but rather fill out more affordable tiers in the new lineup: Sol is designed for complex daily and professional workloads, while Luna targets scenarios where low cost and high throughput are essential.