OpenAI has launched GPT-6 Sol and GPT-6 Luna, extending the GPT-6 generation to the two smaller capability tiers. Each model is reported to cost around half as much as its GPT-5.6 equivalent, a reduction OpenAI attributes to caching and inference improvements.
The timing is worth noting. TechCrunch reports that Anthropic released Claude Opus 5.5 roughly 90 minutes before this announcement, which we cover in our Claude Opus 5.5 article published today. Both vendors have now shipped a cheaper, faster version of an existing generation on the same morning, and both are making claims about accuracy at lower cost. That is the comparison most buyers will actually have to make this week.
The GPT-6 versions of Sol and Luna are not being described here as live in Owlpen today. More on that below. The rest of this article covers what OpenAI has said, what is still unpublished, and what to check before assuming the saving reaches your own invoice.
What has launched
OpenAI separates the generation number from the capability tier. The number identifies the generation, and Sol, Terra, and Luna are durable tiers that can advance on their own cadence, a structure introduced with the GPT-5.6 family and covered in our launch article for that release.
GPT-6 Sol, for complex work
Sol is positioned for complex tasks including coding. OpenAI says it makes about half as many mistakes as its predecessor, reaching what it calls Astra-level reliability at much lower cost, based on its own internal factuality evaluation. It also reports a lower error rate on coding tasks.
GPT-6 Luna, for high-volume clerical work
Luna is positioned for high-volume tasks with a clear goal: summarising documents, extracting information, and answering short questions. This is the tier where unit cost usually dominates the business case, because the work is repetitive and the output is often an input to something else rather than a finished deliverable.
Where this sits against GPT-6 Astra
OpenAI frames the release as extending the benefits of GPT-6 Astra, which shipped earlier this month, by making that capability more efficient and accessible. In its words, Astra introduced a new generation of intelligence and these models extend its benefits. OpenAI also claims the new models outperform Anthropic's Fable and Opus models, a vendor comparison that we would treat as a starting point for testing rather than as a settled result.
What has not been published yet
Launch-day coverage reports the new models at around half the cost of the GPT-5.6 versions of Sol and Luna, but does not set out per-million-token rates, the context window, the knowledge cutoff, or published benchmark scores for either tier. For reference, the current GPT-5.6 rates are $5.00 and $30.00 per million tokens for Sol and $0.20 and $1.20 for Luna. We would wait for OpenAI's own rate card before rebuilding a cost model.
Who can use it
Sol is reported as available in ChatGPT Work and Codex for most paid accounts, and through the OpenAI API. Luna is reported as reaching the desktop app and Free and Go users, with the rollout across the ChatGPT app and website proceeding gradually through launch day. Staged rollouts of this kind mean that what a colleague sees in their account on the morning of release is not a reliable guide to what your own tenancy has been given.
What it means in practice
Halving the cost of the two working tiers, if the published rates confirm it, changes routing economics more than it changes capability. The relevant question for most organisations is not whether to adopt the newest model, but which stages of an existing pipeline can move down a tier without a fall in usable output. Classification, extraction, tagging, and first-pass triage often can. Drafting, analysis, and anything client-facing usually should not move on price alone.
The accuracy claim deserves the same treatment. Halving the error rate on an internal factuality evaluation is a meaningful direction of travel, but a vendor's evaluation set is not your document set. The risk of hallucination falls with better models and does not disappear, and the cost of correcting a confident wrong answer can exceed the token saving many times over. Any tier change should be validated against a representative sample before it reaches production.
There is also a consumption effect to plan for. Cheaper tokens tend to increase usage, particularly in agentic workflows where one instruction can generate a long chain of calls. A price cut without usage controls can leave total spend flat or higher. Budget caps and per-workflow limits remain the effective control.
Finally, where you buy matters. Access purchased through a third-party cloud marketplace is priced by that provider rather than by OpenAI's published rate card, and reductions on the direct API do not necessarily pass through on the same date, or at all.
Governance points to carry over
A cheaper tier does not reduce any obligation attached to the work it performs. Access control, audit trails, data handling, retention terms, and human review requirements apply in the same way at the new rates as at the old ones, and AI governance is not a function of unit price. Where a model is swapped inside an existing workflow, the sensible discipline is to treat it as a change that needs recording, not as a setting.
Two launches on one morning also make a practical point about dependency. Organisations that have hard-coded a single model name into prompts, tooling, or contracts will do this work repeatedly. Building the routing layer so that the model is a configurable choice, with measurement attached, is the part that keeps paying off.
Owlpen and GPT-6 Sol and Luna
GPT-5.6 Sol, Terra, and Luna are available as selectable models within the Owlpen platform, as described when OpenAI cut the GPT-5.6 rates in July. The GPT-6 versions announced today are a separate matter. Coaley Peak intends to make them available once OpenAI API access is confirmed for the relevant deployment channel and production validation is complete.
Where clients ask us to review routing, we look at the work rather than the rate card. We would expect to answer whether output quality holds on the client's own material, whether the reported efficiency gain survives contact with real prompts that were tuned for a previous tier, and whether any change affects data handling terms, before recommending that anything moves.
Owlpen availability
GPT-6 Sol and GPT-6 Luna are not being described here as live in Owlpen today. Coaley Peak intends to make them available once API access is confirmed and production validation is complete. Configured client workflows will not be moved onto a new model without agreement.
Sources used
This article is based on TechCrunch's launch-day report, which quotes OpenAI's announcement on pricing, positioning, availability, and the factuality claim, and on OpenAI's published GPT-5.6 material for the tier structure and the current GPT-5.6 rates. OpenAI's own announcement page for GPT-6 Sol and Luna could not be retrieved directly at the time of writing, so the detail here is attributed to that report rather than to a vendor page, and we will update this article if OpenAI publishes a full rate card or model specifications.
If you would like to discuss model selection, AI cost control, or the Owlpen platform in the context of your own workloads, contact us at enquiries@coaleypeak.co.uk or read more about the Owlpen platform.
Disclaimer. This article is published by Coaley Peak Ltd for general informational purposes only. The views expressed are those of the author, Stephen Grindley, and do not constitute legal, regulatory, financial, or technical advice. Nothing in this article should be relied upon when making procurement, investment, compliance, or technology decisions. References to third-party products, platforms, and companies are for informational purposes only and do not constitute endorsement. Pricing, accuracy, availability, and comparative claims cited are those reported by OpenAI, or by the press source named above, and have not been independently verified by Coaley Peak. At the date of publication, full per-token pricing and model specifications for GPT-6 Sol and GPT-6 Luna had not been confirmed from a vendor page, published rates are subject to change at any time, may differ where access is purchased through a third-party marketplace, and no saving is guaranteed for any particular workload. References to GPT-6 models and Owlpen describe intended Coaley Peak platform availability only and do not imply client access without configuration, review, and applicable commercial agreement. Readers should seek independent professional advice appropriate to their specific circumstances. Information was accurate to the best of the author's knowledge at the date of publication. Coaley Peak Ltd and Stephen Grindley accept no liability for any loss or damage arising from reliance on the contents of this article.