Inference
EN
Alibaba

Qwen3.8-Flash-Next

Qwen3.8-Flash-Next is a reasoning model from Alibaba. 35 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.64.3±3.2Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.55/sMedian 67/s
First tokenSeconds to the first output token.38.1 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.262KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

180B parameters
16-bit
379.0 GB
8-bit
201.8 GB
4-bit
115.6 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.12 counted68.531 / 218
CodingCode writing and repair.6 counted62.157 / 209
ReasoningLogic problems and puzzles.3 counted62.257 / 227
MultimodalTasks with images and text.7 counted65.025 / 84
KnowledgeFacts and expert knowledge.4 counted55.295 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted73.81 / 63
MathMath problems.1 counted59.4—

Results

35 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
RealWorldQAMultimodal88.5%1 / 18
LiveBench Instruction FollowingInstructionAbove fitFar above this model's index.77.1%5 / 66
LiveCodeBench v6Coding91.9%5 / 59
Artificial Analysis GPQA DiamondKnowledge92.3%19 / 199
GDPval-AA normalizedAgentic56.4%13 / 133
ERQAMultimodal72.3%2 / 18
Instruction Following BenchmarkInstruction81.3%8 / 57
CharXiv ReasoningMultimodal90.6%7 / 49
Graduate-Level Google-Proof Q&AKnowledge91.7%21 / 141
LiveBench Agentic CodingAgentic61.6%12 / 66
Artificial Analysis Long Context ReasoningReasoning79.7%45 / 217
Artificial Analysis MMMU-ProMultimodal79.8%24 / 113
Agents' Last ExamAgenticAbove fitFar above this model's index.51.2%5 / 21
Artificial Analysis Humanity's Last ExamKnowledge38.0%55 / 222
GPQA Diamond · BenchLMKnowledge91.7%21 / 84
MathVisionMultimodal90.6%6 / 24
Critical Physics TasksReasoning11.1%61 / 214
SWE-bench ProCoding62.5%24 / 83
AndroidWorldAgentic84.5%3 / 10
Artificial Analysis Terminal-Bench v2.1Agentic86.1%7 / 21
JobBenchAgentic55.7%12 / 36
LiveBench Coding 2.0Coding87.4%18 / 52
LiveBench ReasoningReasoning87.4%28 / 80
Artificial Analysis Tau3-BankingAgentic45.4%8 / 22
CharXiv Reasoning without toolsMultimodal84.6%7 / 17
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal89.9%11 / 26
Artificial Analysis SciCodeCoding50.6%49 / 114
NL2RepoCoding48.1%16 / 36
LiveBench ResearchAgentic74.6%24 / 52
Humanity's Last ExamKnowledge35.9%39 / 77
LiveBench Long AgenticAgentic37.1%26 / 51
Toolathlon-VerifiedAgentic73.5%12 / 23
LiveBench Data AnalysisReasoning74.2%43 / 80
Agent Arena task outcomeAgentic0.331 / 54
Humanity's Last Exam without toolsKnowledge35.9%32 / 53
LiveBench Browser UseAgentic48.0%30 / 49
OSWorld 2.0Agentic19.4%16 / 25
LiveBench DesignAgentic72.8%34 / 52
Agent Arena steerabilityAgentic−2.636 / 54
Artificial Analysis Omniscience AccuracyKnowledge23.5%112 / 164
LiveBench LanguageKnowledge74.6%49 / 66
DeepSWEAgentic58.7%44 / 59
LiveBench MathematicsMath85.8%50 / 66
Agent Arena command recoveryAgentic−2.639 / 50
LiveBench CodingCoding72.6%53 / 66
MathVision with PythonMultimodal95.7%—
LVBenchMultimodal76.6%—
CoWorkBenchAgentic73.9%—
Vision2WebMultimodal64.0%—

Sources

BenchLMCC BY-NC 4.0LiveBenchNo licence statedLMArenaCC BY 4.0

History

5 releases
Index65.4 → 64.3
23 Oct 202612 Nov 2026

Questions

How good is Qwen3.8-Flash-Next?
Its index is 64.3 out of 100. That is place 49 of 283 ranked models.
What is Qwen3.8-Flash-Next best at?
Instruction: 73.8 out of 100, place 1 of 63.
How fast is Qwen3.8-Flash-Next?
55 output tokens per second, 38.1 s to the first token.
What is the context window of Qwen3.8-Flash-Next?
262K tokens.
Is Qwen3.8-Flash-Next open source?
Yes. Its weights are published.
When was Qwen3.8-Flash-Next released?
August 2026.

Badge

Qwen3.8-Flash-Next on Inference360: place 49 of 283, index 64.3
HTML
<a href="https://inference360.ai/models/qwen3-8-flash-next"><img src="https://inference360.ai/badges/qwen3-8-flash-next.svg" alt="Qwen3.8-Flash-Next on Inference360: place 49 of 283, index 64.3" height="20"></a>
Markdown
[![Qwen3.8-Flash-Next on Inference360: place 49 of 283, index 64.3](https://inference360.ai/badges/qwen3-8-flash-next.svg)](https://inference360.ai/models/qwen3-8-flash-next)

Compare with

MiMo-V2.6-Flash64.3Claude Sonnet 564.2Muse Spark 1.164.6GPT-5.464.8Qwen3.8-27B59.2

More from Alibaba

All models
Qwen3.8 FlashUnrankedQwen3.8 Max (0902)63.4Qwen3.8-27B (Chutes TEE)UnrankedQwen3.8 2.4T A95BUnrankedQwen3.8-27B59.2