OpenAI Unveils GPT-5 Tonight: Microsoft Leaks Details Early!

OpenAI hints at GPT-5…

OpenAI has not been subtle. By posting a terse message on X this Wednesday, August 6, stating “LIVE5TREAM THURSDAY 10AM PT (7 PM French time, ed.)“, the company has left little doubt about the upcoming announcement: the launch of GPT-5.

The new flagship model is highly anticipated by enthusiasts. It has been over two years since GPT-4 was released, and although newer, more targeted models have been introduced since, GPT-5 is expected to represent a significant leap forward for the company led by Sam Altman.

This teaser was posted just days after a similarly discreet post by the leader, which displayed a ChatGPT interface with the model selector showing “ChatGPT 5” (see cover image).

…but Microsoft spills the beans

If you thought OpenAI was blunt, wait until you see Microsoft. Although the announcement is only hours away, the Redmond company posted a blog entry on GitHub titled “GPT-5 is now generally available in GitHub Models”, prematurely revealing several key details about GPT-5, including its integration into GitHub Models. The post was quickly removed but captured by several netizens.

The statement highlights notable advances in GPT-5 in terms of reasoning, code generation, and user experience. These improvements are in line with what has been anticipated for several months. Most notably, Microsoft disclosed the existence of four variants of the model:

• gpt-5: designed for logical and multi-step tasks
• gpt-5-mini: a lightweight version for cost-sensitive applications
• gpt-5-nano: optimized for speed, ideal for low-latency uses
• gpt-5-chat: tailored for advanced, natural, multimodal, and contextual conversations in enterprise applications

What to expect from GPT-5?

While GitHub has already revealed several code-related elements, nothing has yet been announced about GPT-5’s other capabilities. However, several prospects are emerging:

• Chain-of-thought reasoning: GPT-5 will natively incorporate step-by-step reasoning, as confirmed by Sam Altman in January. This approach provides more rigorous responses in complex tasks.
• Enhanced memory: the model could retain more information about users’ preferences and habits, enhancing continuity in extended interactions.
• Enhanced multimodality: GPT-5 is expected to further integrate text, image, audio, and video modes to provide a more seamless and contextual experience.
• Agentive dimension: GPT-5 could operate more autonomously within software environments. This direction, hinted at in GitHub’s blog post, would enhance its ability to perform concrete tasks on behalf of the user.

However, OpenAI might also surprise us with the introduction of features that have been kept secret so far. The answers will be revealed this Thursday, August 7, at 7 PM.