Model

Qwen3.5-Omni-Flash

CLOSED

Alibaba · Qwen3.5 family · released Mar 30, 2026

Smaller/faster tier of the Qwen3.5-Omni family (alongside Plus and a third 'Light' tier not in original leads); processes 10+ hours of audio and 400+ seconds of 720p audio-visual input at 1 FPS, with realtime speech generation. Notably released closed-weight, breaking Alibaba's open-weight streak for this class of model per press coverage.

ReasoningCodingVisionFunction callingTool useAgentic

1471.5

Elo · rank #128

Parameters

Undisclosed

Active params

Undisclosed

Context

256K tokens

Architecture

Thinker-Talker omnimodal architecture; both Thinker and Talker adopt Hybrid-Attention MoE; RVQ speech tokenizer (from Qwen3-Omni) plus ARIA (Adaptive Rate Interleave Alignment) text-speech interleaving in the Talker; exact parameter/expert configuration undisclosed

License

Proprietary

Languages

—

API price (in/out)

$0.4 / $2.2

Modalities

text · vision · audio · video

Benchmark results

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

AI2DVision89.0%#22

best: Molmo 72B · 96.3%

C-EvalKnowledge86.0%#24

best: Qwen3.6-Plus · 93.3%

CharXivVision64.4%#26

best: Muse Spark 1.1 (xhigh) · 88.4%

GPQA DiamondReasoning76.4%#95

best: GPT-5.6 · 94.6%

IFBenchReasoning38.4%#51

best: MiniMax M3 · 83.0%

IFEvalReasoning85.2%#58

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding56.6%#107

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

MathVisionVision65.4%#12

best: Seed 2.1 Pro · 92.6%

MathVistaVision82.9%#11

best: Seed 2.1 Pro · 90.7%

MMBench (English)Vision88.8%#8

best: Qwen3.5-397B-A17B · 93.7%

MMLU-ProKnowledge79.9%#60

best: Claude Fable 5 · 91.5%

MMLU-ReduxKnowledge90.0%#21

best: Qwen3.7-Max · 95.0%

MMMU-ProVision68.2%#32

best: Claude Opus 4.7 · 85.5%

MMMUVision76.9%#31

best: Claude Fable 5 · 89.3%

OCRBenchVision89#18

best: InternVL3-78B · 906

RefCOCOVision92.6%#7

best: Qwen3.5-Omni-Plus · 95.0%

SuperGPQAReasoning54.9%#19

best: Qwen3.7-Max · 73.6%

Video-MMEVision77.0%#13

best: Seed 2.1 Pro · 89.2%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.4

Output / M tok

$2.2

Qwen3.5 family

Elo progression across releases

Qwen3.5-397B-A17BFeb 20262182.0 Qwen3.5-PlusFeb 20261882.6 Qwen3.5-122B-A10BFeb 20262021.4 Qwen3.5-Omni-FlashMar 20261471.5 Qwen3.5-Omni-PlusMar 20261834.0

Resources

DOCShttps://qwen.ai/blog?id=qwen3.5-omni APIAvailable via Alibaba API

API price $0.4/$2.2 · each benchmark row carries its own source badge (see methodology)