OpenAI Releases GPT-6 Sol and Luna Across API, Codex and ChatGPT Work
The two lower-cost GPT-6 options reach OpenAI's own developer and work surfaces with different access for paid subscribers and desktop users.
Edited by Tyronne Panaino
OpenAI announced GPT-6 Sol and GPT-6 Luna on September 22, extending the two models across its own API, Codex and ChatGPT Work surfaces. The official OpenAI Developer Community announcement says both models are rolling out to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, while GPT-6 Luna is also available to Free and Go users through the desktop app.
The release matters to developers and teams choosing between capability, speed and cost because OpenAI is presenting Sol and Luna as two more affordable ways to use work derived from the GPT-6 Astra generation. Both models are also available through the API, giving application builders a first-party route to evaluate them outside a subscription interface.
Availability differs by surface
The announcement groups paid ChatGPT Work and Codex access together but gives Luna an additional desktop path for Free and Go users. It does not say that every account receives the models at the same moment, nor does it give a completion time for the rollout. Teams should therefore verify availability in the product and account tier they actually use before treating access as universal.
API availability is the more consequential change for software builders because it allows controlled evaluations inside existing applications and agent workflows. The source does not document migration behavior, context limits, rate limits or service-level characteristics, so those operational questions remain separate from the release announcement itself.
The pricing comparison needs its qualifier
OpenAI says API prices for Sol and Luna are 50% lower than GPT-5.6 promotional pricing. That is a vendor-stated relative comparison, not an independent measurement of total workload cost. The retained announcement does not list absolute input or output token prices, and it provides no evidence about how many tokens a representative task will consume.
For buyers, the practical test is therefore broader than the headline percentage. A useful evaluation would compare task success, retries, latency, tool use and total tokens for the same workload. Lower unit pricing can create more room to iterate, but the source alone cannot establish that a particular production job will cost half as much end to end.
Evidence quality and limitations
This article relies on one first-party OpenAI announcement. It establishes the model names, product surfaces, eligible plan groups and the company's relative-pricing claim. It does not independently validate capability, reliability, safety, efficiency or the completeness of the rollout.
The source describes Sol and Luna as carrying advances from GPT-6 Astra into faster and more affordable models, but it supplies no independently reproduced evidence in the material used here. Internal confidence is medium because the availability announcement is authoritative for OpenAI's own products while the positioning and pricing implications still require real-world testing.
What to watch next
The next verifiable checkpoints are broad account availability, stable API access, published operational limits and workload-level comparisons that include retries and total token use. Those would show whether the models' lower stated API pricing translates into lower costs without sacrificing the reliability a given application needs.
Status
Confirmed. OpenAI has announced the models across its own API, Codex and ChatGPT Work surfaces; performance and end-to-end cost outcomes remain vendor claims pending independent evaluation.
Sources
Update note: Last reviewed 2026-09-23. We will revise this post when OpenAI publishes broader rollout or operational evidence.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.