Artificial intelligence continues to evolve at high speed. And this time, it is not the usual mastodons that lead the dance.
In this video, we come back to three major announcements that could well rebatter the cards in the AI race: Deepseek, Xiaomi and Microsoft each unveiled models with clear ambitions. They bet on logical reasoning, compactness of models and the speed of execution. A trio which directly targets the expert uses of AI, by focusing less on the spectacular demo than on efficiency in real conditions.
Deepseek strikes hard with Prover V2a model focused on mathematical resolution and formal reasoning. For his part, Xiaomi launched Mimo-7ba multilingual model optimized for mobile devices and on -board uses. Finally, Microsoft proposed Phi-4 Reasoningcut for complex tasks and pure logic. Everyone goes there with their strategy, but all target the same goal: to go where GPT-4 still shows limits.
So, should we talk about rocking? Watch the video to have an overview of the performance, technical choices and uses targeted by these new models. The objective: to understand what really changes, beyond the announcements. We continue to follow the AI as close as possible. Remember to subscribe to the chain so as not to miss anything of the next advances and innovations on the tool side.
.