openPangu Pro MoE
Huawei's open 72B mixture-of-experts model, trained entirely on Ascend chips.
About
On June 30, 2025, Huawei open-sourced part of its Pangu family under the openPangu name, including the 72-billion-parameter Pro MoE model with 16 billion active parameters per token. Huawei presents it as the first large-scale model trained entirely on its own Ascend hardware stack, paired with Ascend-optimized inference tooling, as a bid to build an open ecosystem around Chinese accelerator hardware.
One caveat belongs in any writeup: in July 2025 external researchers published analysis alleging strong similarity between Pangu Pro MoE's weights and Alibaba's Qwen-2.5 14B; Huawei's Noah's Ark Lab rejected the plagiarism claim, but the dispute is unresolved. Confirm the openPangu license terms before commercial use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (verified tier, metrics-v1). Evidence: July 2025 external analysis alleged weight similarity to Qwen-2.5 14B; Huawei denied, dispute unresolved.; Ascend-optimized inference stack included in release.. License: Released under Huawei's openPangu license; confirm exact terms before commercial use..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · July 2025 external analysis alleged weight similarity to Qwen-2.5 14B; Huawei denied, dispute unresolved.
- · Ascend-optimized inference stack included in release.