News of the day
1. Deepseek V3.2, an open-source AI model, matches GPT-5 and Gemini 3 Pro on key benchmarks, reaching IMO gold level → Read more
2. Ontology provides essential guardrails for AI agents, ensuring they understand business context and avoid misunderstandings and hallucinations → Read more
3. Runway's Gen-4.5 video model leads benchmarks, surpassing Google & OpenAI, but still struggles with logic errors → Read more
4. HSBC partners with Mistral AI for global generative AI integration, aiming to enhance operations and services worldwide → Read more
Our take
Hi Dotikers!
The AI war between China and the United States just took another turn. And once again, the open source camp is scoring points.
DeepSeek, the Chinese lab that already shook Silicon Valley in early 2025 with its R1 model, is back at it. Their new V3.2 model has just been released, and it directly rivals GPT-5 from OpenAI and Gemini 3 Pro from Google on most performance benchmarks. We're talking about the two most advanced models in the world, developed by companies valued at hundreds of billions of dollars.
But the most impressive part is elsewhere. At the 2025 International Mathematical Olympiad, a competition notoriously difficult for artificial intelligence, DeepSeek won a gold medal. For context: this summer, OpenAI and Google had each announced with great fanfare that they had developed models capable of reaching this level. Many researchers even thought it was impossible for an AI based solely on language. Well, DeepSeek beat them all to the finish line. And the cherry on top: their model is entirely open source and free.
The key takeaway: the gap between open source and proprietary models is shrinking fast. Two years ago, comparing a free model to GPT would have been laughable. Today, DeepSeek is playing in the same league as the American giants. And it's giving away its cards for free.
A.
