Inference
EN
Alibaba

Qwen3.5-27B

Qwen3.5-27B is a reasoning model from Alibaba. 24 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text · image · video
  • Released 2026-03
RankedEnough results for an index.
IndexAll categories, out of 100.51.3±4.3Median 51.0
CoverageIndex weight that has results.90%
SpeedOutput tokens per second.69/sMedian 67/s
First tokenSeconds to the first output token.1.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.26Median $0.60
Output / 1MUS dollars per 1M output tokens.$2.60Median $2.55
ContextMaximum tokens in one request.262KMedian 262K
EloLMArena rating and rank.1409±4 #144Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
qwen/qwen3.5-27b
Max outputMaximum output tokens per request.
82K
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.
—
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.
5 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
ChutesHost$0.20$1.75262K
SiliconFlowHost$0.25$2.00262K
Novita AIHost$0.30$2.40262K
Deep InfraHost$0.26$2.60262K
OpenRouterRouter$0.26$2.60262K

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.6 counted50.3127 / 218
CodingCode writing and repair.2 counted52.7122 / 209
ReasoningLogic problems and puzzles.3 counted50.5110 / 227
MultimodalTasks with images and text.5 counted53.958 / 84
KnowledgeFacts and expert knowledge.5 counted49.0141 / 268
MultilingualTasks in many languages.1 counted51.6—
InstructionTasks with strict prompt rules.2 counted57.120 / 63
MathMath problems.None counted——

Results

24 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Instruction-Following EvalInstruction95.0%4 / 70
Artificial Analysis IFBenchInstruction75.6%21 / 146
τ²-Bench Tool-Agent-User EvaluationAgentic93.9%29 / 149
Massive Multitask Language Understanding ProfessionalKnowledge86.1%15 / 75
Massive Multi-discipline Multimodal UnderstandingMultimodal82.3%8 / 28
Artificial Analysis Long Context ReasoningReasoning77.7%74 / 217
Artificial Analysis GPQA DiamondKnowledge85.8%71 / 199
Graduate-Level Google-Proof Q&AKnowledge85.5%52 / 141
V*Multimodal93.7%6 / 14
Artificial Analysis Humanity's Last ExamKnowledge23.9%101 / 222
LiveCodeBench v6Coding80.7%27 / 59
Artificial Analysis MMMU-ProMultimodal75.0%53 / 113
Software Engineering Benchmark VerifiedCoding72.4%60 / 114
MMLU-ProXMultilingual82.2%9 / 17
LongBench v2Reasoning60.6%14 / 25
Critical Physics TasksReasoning0.9%124 / 214
MathVisionMultimodal86.0%16 / 24
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesKnowledge65.6%15 / 22
Massive Multi-discipline Multimodal Understanding ProMultimodal75.0%38 / 53
CharXiv ReasoningMultimodal79.5%36 / 49
Gert Labs Composite Game BenchmarkAgentic39.4%39 / 52
Artificial Analysis Omniscience AccuracyKnowledge20.7%127 / 164
BrowseCompAgentic61.0%51 / 63
Terminal-Bench 2.0 · BenchLMAgentic41.6%36 / 41
OSWorld-VerifiedAgentic56.2%34 / 37
Artificial Analysis ITBench-AAAgentic35.5%23 / 24
Artificial Analysis Global-MMLU-LiteMultilingual86.8%—
Multimodal Multi-disciplinary Video UnderstandingMultimodal73.3%—
SWE-RebenchCoding58.9%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reusemodels.devMIT

History

5 releases
Index52.1 → 51.3
23 Oct 202612 Nov 2026
Output / 1M$2.60 → $2.60
23 Oct 202612 Nov 2026

Questions

How much does Qwen3.5-27B cost?
$0.26 per 1M input tokens, $2.60 per 1M output tokens.
How good is Qwen3.5-27B?
Its index is 51.3 out of 100. That is place 140 of 283 ranked models.
What is Qwen3.5-27B best at?
Instruction: 57.1 out of 100, place 20 of 63.
How fast is Qwen3.5-27B?
69 output tokens per second, 1.3 s to the first token.
What is the context window of Qwen3.5-27B?
262K tokens.
Is Qwen3.5-27B open source?
Yes. Its weights are published.
When was Qwen3.5-27B released?
March 2026.

Badge

Qwen3.5-27B on Inference360: place 140 of 283, index 51.3
HTML
<a href="https://inference360.ai/models/qwen3-5-27b"><img src="https://inference360.ai/badges/qwen3-5-27b.svg" alt="Qwen3.5-27B on Inference360: place 140 of 283, index 51.3" height="20"></a>
Markdown
[![Qwen3.5-27B on Inference360: place 140 of 283, index 51.3](https://inference360.ai/badges/qwen3-5-27b.svg)](https://inference360.ai/models/qwen3-5-27b)

Compare with

MiniMax M2.751.2GLM-551.6Nemotron 3.5 Super VL 120B A12B51.0o351.8Qwen3.5-122B-A10B51.9

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from Alibaba

All models
Qwen3.5-122B-A10B51.9Qwen3.5-35B-A3B49.4Qwen3.5 FlashUnrankedQwen3.5 PlusUnrankedQwen3.5 9BUnranked