openPangu 2.0 Pro
openPangu 2.0
Open weights trained end to end without a single Nvidia chip. That is the headline and the point.
Updated
The read
openPangu 2.0 Pro is a large sparse model Huawei trained entirely on its own accelerators and then published. As a model it is solid rather than leading. As a demonstration that a full frontier-scale training run can happen outside the Nvidia stack, it matters more than its benchmark scores do, and it comes with a very long context and a permissive-enough licence to build on.
Where it shines
- Trained end to end without Nvidia hardware
- Half a million tokens of context on open weights
- Tuned to run well on hardware most models ignore
Worth knowing first
- Best performance is tied to Huawei's own accelerators
- Behind the leading Chinese open models on general benchmarks
Index score
77.7
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 34 of 53
- Builds
- LLMs
- Category
- Open weights
- Output
- Published weights under Huawei's open licence
- Pricing
- Free tier
- Best for
- Deployments tied to non-Nvidia accelerators
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 05Kimi K3Kimi K3ProvisionalThe largest open-weights model anyone has shipped, and it reads a million tokens without blinking.Open weightsFree tier87.5
- 06GLM-5.2GLM-5.2ProvisionalFrontier-class output, an MIT licence, and a million tokens of context. That combination is the story.Open weightsFree tier87.1
- 12Qwen 3.6Qwen 3.6ProvisionalThe open Qwen release most people should actually run, in sizes that fit real hardware.Open weightsFree tier83.3
