DeepSeek released competitive reasoning models with reportedly low training costs, altering market expectations.
Across 2024–2025, DeepSeek launched models (including R1) that rivaled Western systems on math and code, often with open weights and efficient-training narratives.
Chip and software markets reacted: if the frontier is cheaper to reach, API margins and export policies shift.
The episode underscored multipolarity and how hard it is to control diffusion of post-training techniques.
On 30 September 2026 DeepSeek announced on WeChat, per Reuters and 1 October coverage, a partnership with Huawei to develop and open-source programming infrastructure for Ascend chips, including the TileLang language and compute/communication libraries. The stated goal: an independent GPU software ecosystem that reduces reliance on Nvidia.
Models optimized to think longer at inference defined 2025, with leaps in math contests and coding agents.
The o1 series introduced inference with internal “chain of thought” optimized by RL, excelling at math and code.