System Card← leaderboard
Alibaba

Qwen Flash

#20of 28 · general

Qwen3.8 Flash

125Bparams · MoE
1Mcontext
$0.15 / $0.47per 1M in / out
2026-08-26released
weights ↗
50100Coding 23 · #23 of 28 · 28% of the recipeAgents 33 · #20 of 29 · 38% of the recipeCyber: no readingEngineering: no readingMath: no readingHealth: no readingLegal: no readingProfessional 35 · #15 of 28 · 26% of the recipeWriting: no readingInstruct 55 · #10 of 25 · 45% of the recipeHallucination 10 · #26 of 26 · 37% of the recipeMultimodal 31 · #14 of 24 · 39% of the recipe
Coding23#23
Agents33#20
Cyber—
Engineering—
Math—
Health—
Legal—
Professional35#15
Writing—
Instruct55#10
Hallucination10#26
Multimodal31#14

History

rating1,2001,3001,4001,5001,6001,700Apr 2026Jul 2026Qwen3.5 Flash · $0.065/M in · 2026-02-25 · still served · code/webdev 1,244 ratingQwen3.8 Flash · $0.15/M in · 2026-08-26 · still served · code/webdev 1,633 ratingn3.5 Flash
  • Alibaba
  • older → newer
$ / M input tokens2 of 176 priced models · price is a third-party claim, not a vendor quote

Usage

2.2TtokensAug 28 – Oct 6▼ 5% 7d
600B1.2T1.8T2.4TAug 28Sep 1Oct 1

Geography

China16.4%
United States12.8%
Germany5.7%
Spain3.9%
Brazil3.8%
Japan3.4%
Russia3.0%
Hong Kong2.5%
France2.3%
Singapore2.2%

Measured

≈ imputed · amber dot = lab-self-reported among the sources
GPQA Diamond91.7%#14LLM Stats
HLE37.0%#23Artificial AnalysisLLM Stats
LiveBench76.2%#12LiveBench
SciCode50.6%#23Artificial Analysis
CritPt11.1%#23Artificial Analysis
GDPval56.5%#11Artificial Analysis
AA Index39.8#18Artificial Analysis

Neighbors

MiMo Flash→ looser instructions · weaker multimodalDeepSeek Pro→ more factual · pricierClaude Haiku→ looser instructions · stronger codingGLM Flash→ looser instructions · more factualQwen Lite→ looser instructions · pricier