GitHub accidentally presented GPT-5 the day before the official announcement — 4 different versions of the "most powerful" OpenAI AI are known

GitHub accidentally announced new GPT-5 models before OpenAI.

The post, which was first reported on Reddit, has now been removed, but the information from it available in the archive. From what we know, the new “most powerful” OpenAI AI will offer “significant improvements in reasoning, code quality, and user experience” and will be available in 4 different versions:

• GPT-5 — model for logical and multi-step tasks;
• GPT-5-mini — lightweight version for “cost-sensitive” applications;
• The GPT-5-nano — is optimized for speed and is ideal for low latency applications;
• GPT-5-chat — is designed for rich, natural, multimodal and context-aware conversations for enterprise applications.

OpenAI has “almost” officially confirmed the launch for today, August 7 at 20:00, with a post with a planned “LIVE5TREAM”. CEO Sam Altman specified that today’s live stream “will be longer than usual,” which is expected for four models at once.

Meanwhile, a recent report by The Information lowered expectations for GPT-5. Internal testing has allegedly shown that the model brings improvements in math, coding, and instruction execution, but the performance jump is smaller than what users saw when they moved from GPT-3 to GPT-4. An initial candidate for GPT-5, a large language model codenamed “Orion”did not meet expectations and was instead released as GPT-4.5.

Earlier, the company introduced the first open AI models GPT-OSS, one of which is sufficiently small enough to work locally on a PC.