Huawei open-sources a 505B model trained entirely on its own chips — no Nvidia anywhere
On July 31 Huawei released the weights, inference code and technical report of openPangu-2.0-Pro, a 505-billion-parameter mixture-of-experts model (about 18B active per token) with a 512K context window, trained on roughly 34 trillion tokens. The distinctive part is the hardware: the entire training run happened on Huawei's own Ascend NPUs, with no Nvidia chips involved. The release fulfils a promise Huawei made at its developer conference in June.
Why it mattersUS export controls were meant to keep China from training frontier-scale models — and here is a 505B open model built wholly on domestic silicon, with the recipe published. For the open-source world it is one more heavyweight to build on; for policymakers it is hard evidence that the chip wall is leaking.
Read in the app — free, in 9 languages
Related stories
DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best2026-08-21Show the robot once — three to twelve seconds — and it gets the job right 59 times out of 100 with no training at all
2026-08-21The model invents its own tasks, builds the rig to test them, then trains on the results — and DeepReinforce gave the weights away
2026-08-20A free model that fits on one graphics card scores 52 — then spends 22,276 thinking tokens drawing a picture
2026-08-18Zhipu did not build a new model. It kept training the old one — and says coding got 50 percent better.
2026-08-17