On 30 June 2026 for Flash; 30 July 2026 for Pro, Huawei openPangu announced openPangu 2.0 Flash and Pro. The family arrived in two stages, with a smaller Flash checkpoint followed one month later by the larger Pro model.
This release brief was checked against first-party material on 10 August 2026. The date above is the public announcement date, not the date a repository was created or a third-party provider added the model. Where access or weights arrived later, that distinction is recorded below.
Release record
| Field | Verified detail |
|---|---|
| Announcement | 30 June 2026 for Flash; 30 July 2026 for Pro |
| Availability or weight release | Flash weights on 30 June; Pro weights on 30 July 2026 |
| Release type | staged open-weight foundation-model family release |
| Access | weights under the openPangu Model License 2.0 with Ascend deployment tooling |
| Architecture | 92B-total/6B-active Flash and 505B-total/18B-active Pro MoEs with DSA, SWA and three MTP heads |
| Maximum stated context | 512K tokens |
What changed
openPangu 2.0 is a major non-CUDA-centred open release. Both variants were trained on Ascend hardware over about 34 trillion tokens and use a 1:2 mix of dynamic sparse and sliding-window attention. The result is as much an ecosystem statement as a benchmark release: reproducible deployment depends on the openPangu inference stack and access to compatible NPU infrastructure.
The practical comparison is therefore not simply whether openPangu 2.0 Flash and Pro has the largest headline score. Teams need to ask whether its architecture, access terms, latency, tool behaviour and evaluation setup match the workload they actually intend to run. A model can lead one harness while losing on cost, refusal behaviour, multilingual quality or repeatability in another.
Benchmarks worth retaining
| Evaluation | Reported result | How to read it |
|---|---|---|
| Flash AIME 2026 | 93.3; 98.1 with Python | Thinking-mode mathematical reasoning |
| Pro SWE-Bench Verified | 68.5 | Pro thinking-mode repository result |
| Pro BrowseComp | 65.7 | Agentic browsing result with thinking |
These are release-time results, not independently reproduced guarantees. The licence is custom rather than Apache or MIT, and the first-party evaluations use averaged sampling and tool settings that must be retained with the scores. Scores should remain attached to the disclosed effort setting, agent harness, tool access, timeout, context-management policy and judge model. Moving a number into a procurement sheet without those conditions creates false comparability.
Architecture and access
openPangu 2.0 Flash and Pro is described as 92B-total/6B-active Flash and 505B-total/18B-active Pro MoEs with DSA, SWA and three MTP heads with 512K tokens of stated context. Its access position at verification time is weights under the openPangu Model License 2.0 with Ascend deployment tooling. That wording matters: open weights, source-available weights, an API, a product preview and a research demonstration give adopters very different rights and different levels of reproducibility.
Before deployment, record the exact model identifier or checkpoint, inference stack, quantisation, reasoning setting, region, price schedule and supplier terms. If the release uses a custom licence, read the licence itself rather than relying on the word “open” in launch copy. If it is API-only, preserve the dated documentation and change-notice route because the served snapshot can change without a downloadable artefact.
What an evaluation should test next
For openPangu 2.0 Flash and Pro, a credible internal gate should include:
- a frozen set of representative tasks with pass, fail and abstain criteria;
- a matched baseline using the same tools, timeout, prompt budget and reviewer rubric;
- repeated runs to expose variance rather than reporting a single best attempt;
- latency, token use and total task cost alongside task success;
- adversarial, multilingual and long-context cases relevant to the real deployment; and
- rollback evidence showing the previous model can be restored safely.
The wider model change-control guide explains how to keep model, prompt, tool and corpus changes reconstructable. The AI dependency inventory guide covers the release and supplier records needed after deployment.
AIEngine verdict
openPangu 2.0 deserves separate tracking from community re-uploads. Flash offers the accessible active footprint; Pro adds capability. Both require explicit licence and Ascend-runtime review.
This is a launch assessment, not a certification. Benchmark leadership is useful evidence of where to test; it is not authorization to place the model in a high-impact workflow without domain evaluation, security review and an accountable owner.
Primary sources
Image provenance
Hero image: Huawei openPangu official model card via Hugging Face. The locally served WebP is a crop of the first-party release or model-card asset recorded in the repository provenance manifest.
