Inference
EN
Alibaba

Qwen3.8-27B

Qwen3.8-27B is a reasoning model from Alibaba. 59 benchmarks count toward its index, in 8 categories.

  • Open weights
  • Reasoning
  • Input: text · image · video
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.59.2±2.1Median 51.0
CoverageIndex weight that has results.100%
SpeedOutput tokens per second.113/sMedian 67/s
First tokenSeconds to the first output token.0.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.15Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.88Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1436±5 #88Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
qwen/qwen3.8-27b
Max outputMaximum output tokens per request.
236K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.038
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
14 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
VultrCloud$0.15$1.00262K
ChutesHost$0.039$1.25262K
CerebrasHost$0.99$1.49131K
OpenRouterRouter$0.15$1.881.05M
RegoloHost$0.58$2.42120K
Deep InfraHost$0.20$2.50262K
CoreWeaveHost$0.40$3.00262K
Novita AIHost$0.42$3.001M
NebiusHost$0.45$3.00262K
OVHcloudCloud$0.47$3.19262K
Cloudflare Workers AICloud$0.45$3.20262K
Berget AIHost$0.46$3.48262K
ScalewayCloud$0.684$3.76262K
GroqHost$0.80$4.00131K

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

27.78B parameters
16-bit
59.3 GB
8-bit
32.0 GB
4-bit
18.7 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 24 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.20 counted57.485 / 218
CodingCode writing and repair.14 counted60.369 / 209
ReasoningLogic problems and puzzles.6 counted61.260 / 227
MultimodalTasks with images and text.9 counted63.630 / 84
KnowledgeFacts and expert knowledge.5 counted51.2125 / 268
MultilingualTasks in many languages.1 counted60.6—
InstructionTasks with strict prompt rules.2 counted68.54 / 63
MathMath problems.2 counted56.360 / 168

Results

59 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
EuroEval PolishMultilingual65.4% ±0.55 / 209
EuroEval PortugueseMultilingual68.5% ±0.511 / 182
EuroEval GermanMultilingual57.2% ±0.713 / 191
EuroEval ItalianMultilingual66.2% ±0.514 / 174
EuroEval SpanishMultilingual59.0% ±0.617 / 176
EuroEval SwedishMultilingual64.9% ±0.721 / 195
OSWorld-VerifiedAgentic84.3%4 / 37
EuroEval DutchMultilingual62.1% ±0.521 / 175
EuroEval FrenchMultilingual66.8% ±0.422 / 178
Artificial Analysis Long Context ReasoningReasoning82.0%29 / 217
LiveCodeBench v6Coding90.3%8 / 59
Artificial Analysis GPQA DiamondKnowledge90.5%32 / 199
CharXiv ReasoningMultimodal90.2%8 / 49
SWE-benchCoding86.0% ±1.615 / 87
Artificial Analysis Tau3-BankingAgentic48.0%4 / 22
LiveBench Agentic CodingAgentic61.4%13 / 66
GPQA Diamond · Vals AIKnowledge88.9% ±1.632 / 135
Graduate-Level Google-Proof Q&AKnowledge89.2%34 / 141
Instruction Following BenchmarkInstruction79.5%15 / 57
LiveBench Instruction FollowingInstruction72.7%18 / 66
RealWorldQAMultimodal85.9%5 / 18
GDPval-AA normalizedAgentic45.9%38 / 133
LiveCodeBenchCoding84.0% ±1.041 / 142
Artificial Analysis Humanity's Last ExamKnowledge33.9%67 / 222
MMMU ProMultimodal83.9% ±0.928 / 92
SWE-bench ProCoding61.7%27 / 83
Agents' Last ExamAgenticAbove fitFar above this model's index.42.9%7 / 21
ARC-AGI-1 (semi-private)Reasoning87.5%43 / 126
LiveBench Coding 2.0Coding86.8%19 / 52
MathVisionMultimodal90.0%9 / 24
Critical Physics TasksReasoning5.4%81 / 214
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal69.0%15 / 39
LiveBench Data AnalysisReasoning76.6%31 / 80
ARC-AGI-2 (semi-private)Reasoning42.4%52 / 131
AndroidWorldAgentic81.9%4 / 10
Artificial Analysis MMMU-ProMultimodal76.3%46 / 113
Vibe Code Bench v1.1Coding64.8% ±4.854 / 123
GPQA Diamond · BenchLMKnowledge89.2%37 / 84
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal89.8%12 / 26
MMLU ProKnowledge84.3% ±0.463 / 135
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Medical Long Context Reasoning (MLCR-AA)Reasoning22.8%11 / 22
CharXiv Reasoning without toolsMultimodal83.7%9 / 17
LiveBench DesignAgentic75.9%28 / 52
BabyVisionMultimodal65.7%6 / 11
Terminal-Bench 2.1 · BenchLMAgentic73.0%47 / 86
Terminal-Bench 2.1 · Vals AIAgentic58.4% ±5.341 / 75
ERQAMultimodal65.5%10 / 18
Humanity's Last ExamKnowledge30.8%45 / 77
Artificial Analysis SciCodeCoding46.6%67 / 114
NL2RepoCoding42.3%24 / 36
JobBenchAgentic33.4%24 / 36
LiveBench Long AgenticAgentic27.9%34 / 51
Agent Arena task outcomeAgentic−2.737 / 54
LiveBench CodingCoding75.7%47 / 66
VulcanBench v3Coding82.6%10 / 14
LiveBench ReasoningReasoning80.0%58 / 80
Humanity's Last Exam without toolsKnowledge30.8%39 / 53
Code MigrationCoding14.2% ±4.257 / 77
Agent Arena steerabilityAgentic−6.040 / 54
LiveBench MathematicsMath86.2%49 / 66
LiveBench LanguageKnowledge74.3%50 / 66
LiveBench Browser UseAgentic32.0%38 / 49
IOICoding39.1% ±4.937 / 45
Artificial Analysis AutomationBenchAgentic48.2%19 / 23
Agent Arena command recoveryAgentic−5.942 / 50
ProofBench v1.1Math16.0% ±3.746 / 54
Artificial Analysis DeepsecBench-AAAgentic11.5%18 / 21
LiveBench ResearchAgentic59.4%45 / 52
Artificial Analysis Omniscience AccuracyKnowledge17.3%143 / 164
SkillsBenchCoding38.1% ±4.431 / 35
Artificial Analysis CyberGym-E2E-AAAgentic0.0%17 / 19
Artificial Analysis GDP.pdfAgentic16.4%20 / 22
Artificial Analysis Terminal-Bench v4.0Agentic5.6%21 / 22
DeepSWEAgentic42.2%57 / 59
Artificial Analysis Terminal-Bench v2.1Agentic79.8%21 / 21
Artificial Analysis CWE-Bench-AAAgentic26.7%19 / 19
Bug Hunt BenchCoding15.0%25 / 25
Artificial Analysis Cyber IndexAgentic12.7%19 / 19
MedScribeHealthcare83.8% ±2.041 / 109
Tax Agent BenchFinance30.9% ±3.331 / 82
ARC-AGI-2 (public eval)Reasoning38.8%47 / 122
ARC-AGI-1 (public eval)Reasoning87.5%49 / 116
MortgageTaxFinance64.9% ±0.945 / 97
EMBFinance59.7% ±3.036 / 75
LegalBenchLegal82.4% ±0.572 / 149
Legal Research BenchLegal36.1% ±3.347 / 90
ProgramBenchCoding0.0%34 / 65
TaxEval v2Finance70.8% ±0.986 / 143
CyberBench v1.1Other67.8% ±5.729 / 48
Finance Agent (v2)Finance48.6% ±2.149 / 79
MedCodeHealthcare28.7% ±2.098 / 106
Terminal-Bench-ScienceScience0.0%41 / 44
MathVision with PythonMultimodal94.6%—
OmniDocBench 1.5Multimodal91.1%—
BabyVision with PythonMultimodal85.6%—
CoWorkBenchAgentic70.7%—
WebArena-Verified Browser Agent BenchmarkAgentic64.8%—
Vision2WebMultimodal62.9%—
Artificial Analysis EnterpriseOps-GymAgentic44.2%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseVals AINo licence statedARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedEuroEvalMITLMArenaCC BY 4.0models.devMIT

History

5 releases
Index61.4 → 59.2
23 Oct 202612 Nov 2026
Output / 1M$2.20 → $1.88
23 Oct 202612 Nov 2026

Questions

How much does Qwen3.8-27B cost?
$0.15 per 1M input tokens, $1.88 per 1M output tokens.
How good is Qwen3.8-27B?
Its index is 59.2 out of 100. That is place 77 of 283 ranked models.
What is Qwen3.8-27B best at?
Instruction: 68.5 out of 100, place 4 of 63.
How fast is Qwen3.8-27B?
113 output tokens per second, 0.6 s to the first token.
What is the context window of Qwen3.8-27B?
1.05M tokens.
Is Qwen3.8-27B open source?
Yes. Its weights are published.
When was Qwen3.8-27B released?
August 2026.

Badge

Qwen3.8-27B on Inference360: place 77 of 283, index 59.2
HTML
<a href="https://inference360.ai/models/qwen3-8-27b"><img src="https://inference360.ai/badges/qwen3-8-27b.svg" alt="Qwen3.8-27B on Inference360: place 77 of 283, index 59.2" height="20"></a>
Markdown
[![Qwen3.8-27B on Inference360: place 77 of 283, index 59.2](https://inference360.ai/badges/qwen3-8-27b.svg)](https://inference360.ai/models/qwen3-8-27b)

Compare with

Beam 1.0 501B-A23B FP859.2GPT-5.259.0Apodex 1.158.9Muse Spark59.8dots3-note Preview63.4

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from Alibaba

All models
Qwen3.8 Max65.6Qwen3.8 2.4T A95BUnrankedQwen3.7 FlashUnrankedQwen3.8-27B (Chutes TEE)UnrankedQwen3.8 Max Preview64.2