GitHub has prematurely unveiled OpenAI's next-generation GPT-5 models in a blog post that was swiftly deleted. The post briefly revealed that GPT-5 will be released in four variants, promising significant upgrades in reasoning ability, coding performance, and user experience.

Archived versions of the removed announcement show that GPT-5 features improved agentic capabilities and can complete complex coding tasks with minimal input.

The four model versions include a full-strength GPT-5 for logic and multi-step operations, a lightweight mini version for cost-efficient usage, a nano version optimised for speed and low latency, and a chat-focused variant built for advanced, multimodal enterprise interactions.

Reddit users first spotted the post, including comparisons between GPT-5 and competing models such as Llama 4 Scout and Cohere v2.

The accidental release aligns with growing anticipation after OpenAI insiders hinted at a major release earlier in the week, with CEO Sam Altman teasing the update just days before.

OpenAI is expected to launch GPT-5 later today, officially during a live-streamed event. The company has also introduced two new open-weight GPT-OSS models, including one that can run locally on personal devices.