OpenAI delays debut of its first open-weight model | OpenAI Chat | Chat GPT gratis | OpenAI stock | Turtles AI

OpenAI’s “open-weights” model, originally scheduled for release in June, is delayed to late summer. Sam Altman cited unexpected progress in the research team. Meanwhile, Mistral’s European competitor is launching Magistral, a family of open-source reasoning models.

Key Points:

• OpenAI’s open-weights model will be delayed to late summer.
• Sam Altman cited the delay as additional optimizations and reasoning capabilities.
• Mistral unveiled Magistral Small (24 billion parameters) and Medium, multilingual reasoning models.
• Magistral Small is open-source under the Apache 2.0 license, available on Hugging Face and integrated into Le Chat.

OpenAI CEO Sam Altman announced on X that the company’s first “open-weights” model — meaning with freely accessible weights — was delayed from June to late summer. Altman said the research team has done “something unexpected and surprising” that requires additional time, but promises to be worth the wait. The announcement follows previous plans, announced in March, for an open-weights reasoning model to be distributed to developers via global events. OpenAI aims to offer a system with access to parameters, usable on private hardware, and capable of meeting the needs of customization, privacy, and data sovereignty.

Meanwhile, Mistral, a French startup focused on open-source AI, has launched Magistral, its first family of reasoning models. Magistral comes in two versions: Small (24 billion parameters, open-source under Apache 2.0 license, downloadable from Hugging Face) and Medium (proprietary, for enterprise use). Magistral Small is capable of multilingual thinking, using the chain-of-thought technique in Italian, French, Spanish, Arabic, Chinese, and other languages. Magistral Medium, integrated into Le Chat, promises ten times faster responses through "Think mode" and "Flash Answers".

Mistral wants to position itself as a European alternative to US and Chinese proprietary models, supported by the French government and investors such as a16z, and aims to fill the gap with the performance of reasoning models from brands such as OpenAI, Google and DeepSeek. Magistral Medium excels in mathematical and physical tasks (on benchmarks such as AIME and GPQA), although it does not yet surpass Gemini2.5Pro or Claude Opus4 in all metrics.

The AI ​​market is increasingly moving towards open-source reasoning models, with players such as DeepSeek (with R1), Qwen and Mistral in the forefront. OpenAI is responding cautiously: the postponement seems to be dictated by the desire to avoid a hasty release and instead guarantee concreteness in performance and security.

We just have to wait until the release of OpenAI’s open-weights model this summer to see whether it actually offers superior capabilities to available alternatives.