Qwen3 Omni 30B A3B Instruct

Alibaba Cloud · qwen/qwen3-omni-30b-a3b-instruct

ProviderAlibaba Cloud
Familyqwen3-omni
Typevlm
Statusactive
Released2025-09-20
Updated2025-09-22
Parameters35.3B
Open weightsyes

What it fits on

Ordered by predicted speed. Bandwidth sets decode rate; memory decides whether it runs at all.

Every figure here is computed, not measured. Fit is weights at each quantisation against device memory, with a 25% allowance for the KV cache, activations and the OS. Decode rate is the memory-bandwidth roofline at 70% efficiency. Nobody has run this model on these devices.
DeviceMemoryBandwidthBest qualityWeightstok/sFastest
Cerebras WSE-344.0 GB21000000.0 GB/sq626.44 GB~555873.5~833810.3 q4
NVIDIA GB200 Grace Blackwell Superchip372.0 GB16000.0 GB/sbf1670.52 GB~158.8~635.3 q4
AMD Instinct MI355X288.0 GB8000.0 GB/sbf1670.52 GB~79.4~317.6 q4
NVIDIA B200180.0 GB7700.0 GB/sbf1670.52 GB~76.4~305.7 q4
Google TPU7x (Ironwood)192.0 GB7380.0 GB/sbf1670.52 GB~73.3~293.0 q4
AMD Instinct MI325X256.0 GB6000.0 GB/sbf1670.52 GB~59.6~238.2 q4
AMD Instinct MI300X192.0 GB5300.0 GB/sbf1670.52 GB~52.6~210.4 q4
NVIDIA H200 SXM141.0 GB4800.0 GB/sbf1670.52 GB~47.6~190.6 q4
NVIDIA H100 NVL94.0 GB3900.0 GB/sfp835.26 GB~77.4~154.9 q4
NVIDIA H100 SXM80.0 GB3350.0 GB/sfp835.26 GB~66.5~133.0 q4
AMD Instinct MI250X128.0 GB3200.0 GB/sbf1670.52 GB~31.8~127.1 q4
Google TPU v5p95.0 GB2765.0 GB/sbf1670.52 GB~27.4~109.8 q4
NVIDIA A100 80GB SXM80.0 GB2039.0 GB/sfp835.26 GB~40.5~81.0 q4
NVIDIA H100 PCIe80.0 GB2000.0 GB/sfp835.26 GB~39.7~79.4 q4
NVIDIA GeForce RTX 509032.0 GB1792.0 GB/sq522.04 GB~56.9~71.2 q4
Google TPU v6e (Trillium)32.0 GB1638.0 GB/sq522.04 GB~52.0~65.0 q4
AMD Instinct MI21064.0 GB1600.0 GB/sfp835.26 GB~31.8~63.5 q4
NVIDIA A100 40GB SXM40.0 GB1555.0 GB/sq626.44 GB~41.2~61.7 q4
Google TPU v432.0 GB1200.0 GB/sq522.04 GB~38.1~47.6 q4
NVIDIA GeForce RTX 409024.0 GB1008.0 GB/sq417.63 GB~40.0~40.0 q4
AMD Radeon RX 7900 XTX24.0 GB960.0 GB/sq417.63 GB~38.1~38.1 q4
NVIDIA GeForce RTX 309024.0 GB936.0 GB/sq417.63 GB~37.2~37.2 q4
NVIDIA L40S48.0 GB864.0 GB/sfp835.26 GB~17.2~34.3 q4
Apple M3 Ultra512.0 GB819.0 GB/sbf1670.52 GB~8.1~32.5 q4
Apple M2 Ultra192.0 GB800.0 GB/sbf1670.52 GB~7.9~31.8 q4
Apple M4 Max128.0 GB546.0 GB/sbf1670.52 GB~5.4~21.7 q4
Apple M1 Max64.0 GB400.0 GB/sfp835.26 GB~7.9~15.9 q4
Apple M2 Max96.0 GB400.0 GB/sbf1670.52 GB~4.0~15.9 q4
Apple M3 Max128.0 GB400.0 GB/sbf1670.52 GB~4.0~15.9 q4
NVIDIA L424.0 GB300.0 GB/sq417.63 GB~11.9~11.9 q4
Apple M4 Pro64.0 GB273.0 GB/sfp835.26 GB~5.4~10.8 q4
Apple M432.0 GB120.0 GB/sq522.04 GB~3.8~4.8 q4

Where it runs

Taken from the card's availability section.

PlatformId on that platform
Hugging FaceQwen/Qwen3-Omni-30B-A3B-Instruct

Verified benchmark evidence

Each row was checked against its source by a reviewer, and carries the model identifier as actually evaluated — which is not always the same as this card's.
BenchmarkEvaluated asScoreEvidence dateSource kind
aa_lcrQwen3 Omni 30B A3B0.0%2026-09-10 evaluatedindependent evaluatorsource
critptQwen3 Omni 30B A3B0.0%2026-09-10 evaluatedindependent evaluatorsource
gpqa_diamondQwen3 Omni 30B A3B62.02%2026-09-10 evaluatedindependent evaluatorsource

Reported benchmark scores

This card reports no benchmark scores yet.

BenchmarkCatalogue standingScore

Data

This card as JSON · See it in the graph · Edit on GitHub