OpenAI expanded its GPT-6 lineup on September 22 with Sol and Luna, two models positioned as lower-cost options for professional work. Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50, respectively, according to launch coverage from Reuters and VentureBeat.
The release adds two price points below flagship GPT-6 Astra. In its announcement, OpenAI describes Sol and Luna as bringing frontier intelligence to everyday work with different balances of capability and cost. That is the company’s positioning; the supplied evidence does not establish how either model performs on a particular developer’s workload.
API pricing: a substantial gap between Sol and Luna
| Model | Input price per million tokens | Output price per million tokens |
|---|---|---|
| GPT-6 Sol | $2.00 | $10.00 |
| GPT-6 Luna | $0.10 | $0.50 |
At those reported rates, Sol costs 20 times as much as Luna for either input or output tokens. The difference makes model choice consequential for applications that repeatedly call a model, although token prices alone do not establish which option delivers better results for the money.
VentureBeat compared the new rates with GPT-5.6 Sol at $4 per million input tokens and $20 per million output tokens, making GPT-6 Sol 50% cheaper in both directions. Its reported comparison for Luna was $0.20 for input and $1.20 for output, corresponding to reductions of 50% and about 58%, respectively. Reuters also reported the 50% Sol reduction, describing the predecessor’s rates as promotional.
The distinction between earlier promotional pricing and the new rates matters. VentureBeat reported that an OpenAI spokesperson confirmed the GPT-6 Sol and Luna prices were permanent rather than introductory. OpenAI attributes the lower prices to improvements in inference and caching, saying those changes allow it to reduce cost while increasing capability. That explanation is a company claim, rather than an independently measured efficiency result.
Sol targets recurring technical work
VentureBeat describes Sol’s intended work as feature development, code review, debugging and data analysis: complex tasks that developers and knowledge workers perform repeatedly. Reuters reports that OpenAI designed both Sol and Luna as cheaper options for professional work, coding, automation and computer-use tasks.
Astra retains the top position for demanding work where capability takes priority over cost, according to the launch coverage. Sol and Luna therefore broaden the available choices rather than displacing the flagship in the company’s positioning.
OpenAI also told Reuters that it trained the two models using methods similar to those used for Astra, with improvements in reasoning, factual reliability, coding, computer use and alignment. Those statements describe intended improvements, but the research package does not supply benchmark scores, task-level comparisons or latency measurements that would quantify the trade-offs among the three models.
For engineering teams, the practical question is whether a cheaper model completes the required task reliably enough. The reported pricing establishes a clear cost difference; the launch claims do not establish that Luna can substitute for Sol, or that either can substitute for Astra, across every workflow.
Availability was staged at launch
VentureBeat lists the API model identifiers as gpt-6-sol and gpt-6-luna. In its September 22 coverage, it also reported that OpenAI was rolling out both models to ChatGPT Work and Codex for Plus, Pro, Business and Enterprise customers, while Free and Go users could access Luna through the desktop app.
The same report said the models were not yet available in Chat itself and that the Work and Codex rollout would proceed gradually during the day to preserve service stability. Those details describe launch-day availability; they do not establish the rollout’s status at a later date.
For developers evaluating the release, the clearest established change is the price structure: a mid-priced Sol option and a much cheaper Luna option alongside Astra. Whether those savings translate into lower operating costs depends on the quality each model delivers on the work an application actually needs to complete.
