Model

openPangu-2.0-Flash (Non-Thinking)

OPEN

Huawei · openPangu family · released Jun 12, 2026

Default (no-extended-reasoning) inference mode of the same openPangu-2.0-Flash weights; this is the mode a user gets without opting into Thinking mode.

ReasoningCodingVisionFunction callingTool useAgentic

1603.9

Elo · rank #117

Parameters

92B

Active params

6B (MoE)

Context

512K tokens

Architecture

MoE with MLA + DSA/SWA hybrid attention (1:2 ratio), 4-stream mHC residual, multi-token prediction heads; trained on Huawei Ascend NPUs (no NVIDIA hardware)

License

OpenPangu Model License Agreement v2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath86.5%#63

best: GPT-5.2 · 100.0%

GPQA DiamondReasoning79.8%#88

best: GPT-6 Astra · 96.0%

IFBenchReasoning54.4%#38

best: MiniMax M3 · 83.0%

IFEvalReasoning89.3%#25

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding50.9%#119

best: DeepSeek-V4-Pro (Think Max) · 93.5%

PinchBenchAgents82.5%#6

best: Trinity-Large-Thinking · 91.9%

SWE-bench VerifiedCoding57.6%#87

best: Claude Opus 5 · 96.0%

Run it locally

VRAM @ Q4

—

VRAM @ FP16

—

Fits on (Q4)

Multi-node cluster required

Throughput data unavailable.

Quantizations

—

Fine-tune it

Conditional / custom

QLoRA62.6 GB1× A100 80GB

LoRA196.0 GB2× H200 141GB

Full fine-tune1476.6 GB8× B200 192GB

QLoRA SFT on ~10k samples ≈ $4.22 (1× A100 80GB)

Plan a fine-tune →

openPangu family

Elo progression across releases

openPangu-2.0-Flash (Non-Thinking)Jun 20261603.9 openPangu-2.0-Flash (Thinking)Jun 20262057.2

Resources

HFopenpangu/openPangu-2.0-Flash DOCShttps://huggingface.co/openpangu/openPangu-2.0-Flash

API price weights · each benchmark row carries its own source badge (see methodology)