AI BriefingHuawei Launches openPangu 2.0
Key point
Huawei has unveiled openPangu 2.0, a large language model optimized for HarmonyOS, and will make it open source starting June 30.
Details
Huawei officially announced openPangu 2.0 at HDC 2026. The model is fully optimized for the HarmonyOS ecosystem and achieves maximized performance on Huawei's Ascend computing power.
The model is offered in two versions:
- openPangu 2.0 Pro: 505B total parameters (18B active parameters), applying a high sparsity ratio of 28:1
- openPangu 2.0 Flash: 92B total parameters (6B active parameters)
Key technical features include the following:
- Support for 512K context processing capability
- Introduction of a DSA+SWA independent-layer hybrid architecture (Ultra-sparse attention)
- Through Ascend-native training, a 30% improvement in training efficiency and a 50% increase in throughput for 512K long-sequence training
Starting June 30, Huawei plans to progressively open source core components—including model architecture, weights, technical reports, and inference code—followed by pre-training and post-training code.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Qwen3.5-based Small Decision Model 'Kev' Released with TypeSafe API Compatibility and Performance MetricsJevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs Detection