OpenAI has expanded its GPT-6 model family with the launch of GPT-6 Sol and GPT-6 Luna. Both models bring features from the flagship GPT-6 Astra model to cheaper and faster AI systems. The company said the new models extend Astra’s improvements in coding, factual accuracy, computer use and professional workloads while making them more accessible.
GPT-6 Sol targets complex tasks such as coding and AI agents. Meanwhile, GPT-6 Luna focuses on high-volume tasks, including document summarization, information extraction and quick responses. The launch comes shortly after OpenAI introduced GPT-6 Astra earlier this month.
50% Lower API Pricing
OpenAI highlighted efficiency improvements in caching and inference as the reason behind the major price cuts. Therefore, both new models cost significantly less to run than their predecessors.
| Model | Input Price (per 1M tokens) | Output Price (per 1M tokens) |
|---|---|---|
| GPT-6 Sol | $2 | $10 |
| GPT-5.6 Sol | $4 | $20 |
| GPT-6 Luna | $0.10 | $0.50 |
| GPT-5.6 Luna | $0.20 | $1.20 |
The company said GPT-6 Sol and Luna are available at half the cost of their GPT-5.6 counterparts.
Coding and Knowledge Work Improvements
OpenAI said GPT-6 Sol delivers improvements in coding performance, especially for AI agents working on real software projects. Additionally, the company shared benchmark results to support these claims.
| Benchmark | GPT-6 Sol Result | Comparison |
|---|---|---|
| FrontierCode 1.1 Main | Improved over GPT-5.6 Sol | Comparable to Claude Fable 5.1 at lower cost |
| AutomationBench | 33.2% (xhigh effort) | Higher than Claude Opus 5 (26.9%) |
| OSWorld 2.0 Offline | 60.5% | Similar to Claude Opus 5 medium effort (60.3%) |
On the FrontierCode 1.1 benchmark, which evaluates whether AI-generated code is suitable for merging into real codebases, OpenAI said GPT-6 Sol improved significantly over GPT-5.6 Sol. The company also said it matched Claude Fable 5.1 at a lower cost.
Better Factual Accuracy
OpenAI said GPT-6 Sol produces fewer factual errors compared with its predecessor. According to the company’s internal evaluation, based on real-world conversations where users reported mistakes, GPT-6 Sol made around half as many errors as GPT-5.6 Sol. Still, this figure comes from OpenAI’s own internal testing.
Computer Use and AI Agents
OpenAI said GPT-6 Astra remains its most capable model for computer-use tasks. However, GPT-6 Sol and Luna offer improved cost efficiency for similar work. On the OSWorld 2.0 benchmark, OpenAI reported that GPT-6 Sol achieved similar performance to Claude Opus 5 while using a significantly lower cost per task. Furthermore, GPT-6 Luna improved over previous-generation models at a lower operating cost.
GPT-6 Sol vs GPT-6 Luna
| Feature | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| Target Users | Developers, professionals, AI agents | High-volume everyday tasks |
| Main Focus | Coding, reasoning, complex workflows | Summaries, extraction, quick answers |
| Performance Level | Higher capability | Lower cost and faster execution |
| Best Use Cases | Software development, automation | Business workflows and routine tasks |
Availability
GPT-6 Sol and Luna are now available through ChatGPT Work, Codex and the OpenAI API. Also, OpenAI said Luna will become available for Free and Go users through the ChatGPT desktop app. Meanwhile, the broader ChatGPT rollout will continue gradually. Finally, both models are accessible through the API as gpt-6-sol and gpt-6-luna.
