Model

Qwen3.8-Omni-Flash

CLOSED

Alibaba · Qwen3.8 family · released Sep 18, 2026

Alibaba's next-generation native omnimodal model (text, image, audio and video in; text out), launched on 2026-09-18 as an API-only model with a 1M-token context, 4x the 256K of Qwen3.5-Omni-Flash. Qwen says its text performance matches a text-only model of the same size: its text table lines up closely with Qwen3.8-Flash (SWE-bench Pro 63.3 vs 62.5, GPQA 91.0 vs 91.7). Its omnimodal and agentic rows improve sharply over Qwen3.5-Omni-Plus, for example WildClawBench-MM rises from 34.5 to 71.0 and StreamingBench from 57.1 to 80.8. International pricing is $0.15/$0.47 per Mtok, and parameter count is undisclosed.

ReasoningCodingVisionFunction callingTool useAgentic

2531.5

Elo · rank #39

Parameters

Undisclosed

Active params

—

Context

1M tokens

Architecture

Proprietary model; architecture and parameter count undisclosed (API-only); natively omnimodal text/image/audio/video input, 1M-token context

License

Proprietary (API only; no weights)

Languages

—

API price (in/out)

$0.15 / $0.47

Modalities

text · vision · audio · video

Benchmark results

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

AndroidWorldAgents87.1%#1

best: this model · 87.1%

CharXivVision83.5%#10

best: Qwen3.8-Flash-Next · 90.6%

DeepSWECoding57.8%#23

best: Gemini 4 Argon · 77.9%

GPQA DiamondReasoning91.0%#26

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning36.5%#41

best: Claude Opus 5.5 · 67.7%

IFBenchReasoning81.5%#5

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding92.6%#2

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

LVBenchVision76.9%#6

best: Gemini 4 Argon · 91.7%

MathVisionVision91.8%#2

best: Seed 2.1 Pro · 92.6%

NL2RepoCoding48.9%#5

best: DeepSeek-V4.1-Flash · 64.0%

StreamingBenchVision80.8%#1

best: this model · 80.8%

SWE-bench MultilingualCoding80.5%#7

best: Claude Opus 5.5 · 93.9%

SWE-bench ProCoding63.3%#11

best: Claude Opus 5.5 · 89.9%

Run it locally

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

Input / M tok

$0.15

Output / M tok

$0.47

Qwen3.8 family

Elo progression across releases

Qwen3.8 MaxAug 20262722.8 Qwen3.8-27BAug 20262458.7 Qwen3.8-Flash-NextAug 20262566.2 Qwen3.8-Max-0902Sep 20262988.4 Qwen3.8-Omni-FlashSep 20262531.5

Resources

DOCShttps://www.alibabacloud.com/help/en/model-studio/qwen3-8-omni-flash APIAvailable via Alibaba API

API price $0.15/$0.47 · each benchmark row carries its own source badge (see methodology)