OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes

Earlier this month, OpenAI launched GPT-6 Astra, which it heralded arsenic its astir almighty and susceptible exemplary yet, and the “world’s champion model” for a assortment of activities, including machine enactment and coding. Now, the laboratory is expanding the GPT-6 procreation with updated versions of the smaller Sol and Luna models.

“GPT-6 Astra introduced a caller procreation of intelligence; these models widen its benefits by making that quality much businesslike and accessible,” the institution writes.

The archetypal models successful the Sol and Luna bid were introduced earlier this year, representing antithetic tiers successful OpenAI’s hierarchy of models. Sol is designed for analyzable tasks similar coding, portion OpenAI says Luna is much due for clerical enactment — “high-volume tasks with a wide goal, similar summarizing documents, extracting information, oregon answering speedy questions.”

OpenAI is besides emphasizing large improvements successful ratio and affordability — astir notably, via a important terms driblet for API access. The 6 bid models volition beryllium disposable astatine fractional the outgo of the 5.6 bid of Sol and Luna, a terms driblet that OpenAI attributes to improvements successful caching and inference.

OpenAI besides claims that its newest models are amended astatine producing answers that are factually accurate, and notes that they person a little mistake complaint erstwhile it comes to coding.

“On our interior factuality evaluation, which is based connected de-identified real-world conversations wherever users flagged mistakes by our models, GPT-6 Sol makes astir fractional arsenic galore mistakes arsenic its predecessor, reaching Astra-level reliability astatine overmuch little cost,” the announcement reads.

As is emblematic with OpenAI’s caller exemplary releases, the institution is claiming that its newest offerings outshine those of its superior competitor, Anthropic. The institution repeatedly claims that GPT-6 Sol and Luna grip tasks substantially amended than Anthropic’s apical models — similar Fable and Opus.

Notably, Anthropic released a caller mentation of Opus 5.5 conscionable 90 minutes earlier OpenAI’s release, reflecting the aggravated contention betwixt the 2 companies.

The caller versions of Sol and Luna are present disposable successful ChatGPT Work and Codex for astir paid accounts, and successful the ChatGPT API, portion Luna volition beryllium disposable successful the desktop app and for Free and Go users. OpenAI expects to rotation the models retired to ChatGPT (the app and website) gradually passim the day.