On September 23, 2026, OpenAI unveiled updated versions of its sixth-generation models — GPT-6 Sol and GPT-6 Luna. According to the developers, both models already occupy their niches within the company's product lineup and are available to users on the day of the announcement. The key focus of the presentation was the quality-to-cost ratio: the updated Sol, according to OpenAI data, makes almost half as many errors as its predecessor, reaching the reliability level of the flagship GPT-6 Astra at a significantly lower price. At the same time, the company substantially reduced API access costs — sixth-series models are now offered at half the price of the 5.6 series.
Role Division: Sol for Complex Tasks, Luna for Routine Work
In the updated lineup, OpenAI clearly divided the functions of the two models. GPT-6 Sol is oriented toward complex tasks and software code writing — it is the workhorse for developers and engineers who need maximum accuracy in non-trivial scenarios. GPT-6 Luna, by contrast, is tuned for routine operations: document summarization, information extraction, and quick answers to standard queries. This positioning allows the company to cover the full spectrum of needs — from high-load computational tasks to mass user traffic — without having to route every query through the most expensive model.
Cost Reduction as a Result of Technological Progress
OpenAI attributes the API price drop to technological progress: optimization of computations and architectural solutions allowed the company to reduce inference costs, which it passed on to customers. Formally, this means that developers who have integrated sixth-series models into their products get access at half the price compared to the 5.6 series, with — as OpenAI claims — no lower, and in the case of Sol even higher, quality. For the market, this is the classic "price falls, quality rises" scenario that has historically accompanied transitions between generations of large language models.
Race with Anthropic: Release 90 Minutes Before the Announcement
The presentation of OpenAI's updated models once again underscored the fierce competition with Anthropic. Ninety minutes before the release of Sol and Luna, Anthropic unveiled an updated version of its Opus 5.5 model. The close timing of the two industry giants' releases sent a clear signal that competition in the frontier-model segment has entered a phase of nearly synchronized updates. Meanwhile, OpenAI publicly claims that the new Sol and Luna outperform rivals' top solutions, specifically the Fable and Opus models. These statements are one side's position and are not supported by independent benchmarks in the provided materials.
Availability: ChatGPT Work, Codex, API, and the Free Tier
The updated Sol and Luna are already available in ChatGPT Work and Codex for most paid accounts, as well as via the API. The Luna model will additionally appear in the desktop application and become available to free-tier and Go-tier users — an important step for the mass segment, since Luna's routine scenarios (summarization, information extraction, quick answers) are precisely the foundation of everyday use. Thus, OpenAI is effectively extending the reach of its updates to free users, using Luna as a "gateway" into the sixth-generation ecosystem.
