OpenAI has expanded the GPT‑6 lineup with two models, Sol and Luna. They are available in the API, Codex and ChatGPT Work and are built for different jobs: Sol for more complex work with code and agents, Luna for fast, low-cost operations.
How Sol differs from Luna
GPT‑6 Sol is aimed at complex multi-step tasks: programming, working with tools and agentic scenarios. In the API it supports images, structured outputs, function calling and built-in tools.
GPT‑6 Luna is designed for cases where speed and cost matter more: bulk processing, short requests, drafts and supporting stages of a larger process. It is not a “lite button” for every task but a different balance of resources.
What changed in access
Both models are available in the Work and Codex environments and through the API. In the regular ChatGPT chat the set of models may differ: it depends on the plan, the country, the workspace settings and the rollout stage.
In the API, OpenAI lists different prices: for Sol, $2 per million input tokens and $10 per million output tokens; for Luna, $0.10 and $0.50 respectively. Price is not the only criterion: answer quality, context size and the need for extra tools matter too.
How to choose a model
For complex debugging, architecture design or an agent working through several steps, it makes sense to start with Sol. For classification, data extraction, short summaries and bulk tasks, it is often wiser to try Luna.
The best way to compare the models is to take a few of your typical tasks, define quality criteria in advance and test the same inputs. Then your choice rests on results, not on the model’s name.
Leave a Reply