DeepSeek has just made a move that could change the ecosystem of reasoning models: the launch of DeepSeek-Math-V2.

Chinese startup DeepSeek has developed a new open-weight AI model, Math-V2, capable of generating and self-verifying complex mathematical theorems.

According to DeepSeek, the newly updated “regular lineup” V3.2—now available on the web, app, and via API—strikes a balance between reasoning ability and output length, making it well-suited for ...

DeepSeek had been somewhat under the radar since releasing its V3 and R1 models that first put China’s AI capabilities in focus, ...

In July, an advanced version of Google DeepMind’s Gemini model and an experimental reasoning model from OpenAI also achieved the gold status on the IMO 2025. Like DeepSeek’s new model, both OpenAI and ...

China’s DeepSeek has released the world’s first open-source AI model capable of scoring at gold medal level in the ...

DeepSeek-Math-V2 is said to match the performances of OpenAI and Google DeepMind’s models on problems from the International Maths Olympiad 2025.

Deepseek version 3.2 packs 671B parameters with 37B active at inference, giving you faster tool use and lower run costs on ...

The Hangzhou-based company released DeepSeek-V3.2 and a more specialized variant, DeepSeek-V3.2-Speciale, which it says can ...

DeepSeek 3.2 introduces DeepSeek Sparse Attention for long context, improving tool use while keeping memory across calls, finish multi-step tasks faster ...

Chinese open-source models are hot on the heels of their US counterparts from frontier labs. DeepSeek has launched two new ...