AI
OpenAI rolls out cheaper, more accurate versions of GPT-6 Sol and Luna models
Following its flagship Astra release, OpenAI is bringing faster, lower-cost updates to its smaller model tiers with claimed gains in factual accuracy and coding performance.
TechCrunch — AI
OpenAI has introduced refreshed iterations of its GPT-6 Sol and Luna models, designed to deliver the capabilities of the earlier Astra release at improved efficiency and reduced expense.
Expanding the GPT-6 lineup#
The new Sol and Luna editions follow OpenAI's recent unveiling of GPT-6 Astra, which the company positioned as its most capable model to date. According to OpenAI's announcement, the updated smaller models "extend its benefits by making that intelligence more efficient and accessible." Sol targets demanding workloads such as software development, while Luna handles high-volume, structured assignments like document summarization and data extraction.
Major price reductions and performance gains#
The most significant change is a 50% price cut for API access compared to the earlier 5.6-generation versions of both models. OpenAI credits better caching techniques and inference optimization for this reduction.
Beyond cost, the company highlights measurable reliability improvements:
- Sol's accuracy: OpenAI reports that GPT-6 Sol now produces roughly half the errors of its predecessor on internal factuality assessments, matching Astra-level dependability
- Coding performance: Both models show reduced error rates in code generation tasks
- Factual grounding: Improvements drawn from real-world usage data where users flagged inaccuracies
Availability and rollout#
The models are now accessible through:
- ChatGPT Work and Codex (for most paid subscribers)
- ChatGPT API
- Desktop applications and Free/Go tier access (Luna)
OpenAI expects to roll out general availability to the main ChatGPT platform throughout the day.
Competitive timing#
The launch underscores ongoing rivalry with Anthropic. OpenAI asserts that Sol and Luna substantially outperform Anthropic's leading models, including Fable and Opus. Notably, Anthropic released its own update—Opus 5.5—just 90 minutes before this announcement, highlighting the rapid innovation pace between the two firms.
