A Chinese AI model developer. It stunned the world by building top-tier models at a fraction of the cost, and it releases its model weights in full. Its flagship model is DeepSeek-V4-Pro. Officially launched in August 2026, V4-Pro comes with a feature that lets users adjust reasoning intensity across three levels. That same month, OpenAI and Anthropic rolled out lower-priced new products, a move widely seen as a response to the wave of cheap Chinese models. Meanwhile, a jailbreak prompt originally made for Google's Gemma was reported in August 2026 to also work on DeepSeek V4 Flash, and around the same time, the DeepSeek-V4-Flash-0731 model was flagged for freezing during long-context tasks.
Current rank (1M)
15
Rank over the last 30 days
2026.08.20 – 2026.09.18 · High 6 · Low 15 · Now 15
The Tencent Hunyuan team released the EvolveScaler paper and project page on September 15, an information-evolution data framework. On data built by defining state in code first and then rendering it as natural language, the median avg@5 across 14 models fell to 11.3 on the longest tier.
Chinese President Xi Jinping said at the 18th BRICS Summit in New Delhi on September 13 that China would take the lead in building a BRICS AI open-source community. It topped a list of five cooperation initiatives, and the intelligent-manufacturing item included the words standards and norms.
Cognition's SWE-2 takes China's Kimi K3 and runs it through another round of reinforcement learning. The company says it lands within a point of Fable 5.1 while costing 64% less.
Anthropic published a 154-page report exposing unauthorized distillation by seven Chinese AI labs. But as the company itself admits, distillation is legal — the real problem isn't the stolen accounts, it's your conversations.
Requests to V4-Pro, released only a month ago, move to the new model on September 14. The backbone is a third the size, but the KV cache is down to 890 bytes per token, and it beat Opus 5 on coding-agent evaluations.
Two hours by train from Beijing, roughly 100 data centers have sprung up in Ulanqab, Inner Mongolia. Cheap electricity, cold winters, and surplus wind power have drawn China's AI companies to this plateau — but underneath sit two unresolved problems: water scarcity and coal. One city tells you how China's AI infrastructure math actually works.
Tencent says the model beat GPT-5.6 on a coding evaluation and even helped optimize its own training and inference process
Z.ai's IPO gong, an order system rebuilt on Kimi K3, free DeepSeek access, and a 300-person WAIC party: founders, developers, and investors keep converging on this one spot, turning it into lived-in infrastructure for China's AI ecosystem.
Taiwan's TeamT5 says DeepSeek, ChatGPT, and Claude Code were used in real-world attacks
'Transfer station' proxies that defeat even selfie verification are undermining regional blocks and identity checks
Community tokenizer comparison shows 60/60 match with zero average error — but that doesn't confirm it's the same model
Waveguide glasses with no speaker or camera that only display text, with a subscription adding ChatGPT and Gemini on top
