Skip to content

Qwen3.5 397B

AlibabaOpen weightsNon-reasoningProvisionalReleased Feb 16, 2026
Compare with GPT-5.2-CodexSource scores ↗
Overall score
55.4
Rank 66 of 222 · 41.1–69.8
Context window
128K
128,000 tokens
Price per 1M tokens
$0.60 in · $3.60 out
Source: BenchLM
Blended price
$1.35
3:1 input:output tokens
Output speed
—
Not measured yet

Category scores

Scores per capability area with the model's rank among ranked models in that category.

  1. 1Multilingual#593.8
  2. 2Math81.7
  3. 3Instruction following#5979.9
  4. 4Reasoning74.1
  5. 5Multimodal#2973.6
  6. 6Knowledge#7061.2
  7. 7Coding41.7
  8. 8Agentic & tool use#6239.9
Higher is better. #N is the model's rank in that category, where it is eligible to be ranked.

Benchmark results (61)

As reported in system cards, papers and public evaluations. Position compares this model with every other model reporting the same benchmark.

Multimodal

MMMU-Pro79.0#15 of 48
CharXiv80.8#26 of 42
MMMU85.0#4 of 21
MathVision88.6#8 of 20
SimpleVQA67.1#10 of 17
RealWorldQA83.9#8 of 16
ERQA67.5#8 of 16
ScreenSpot Pro65.6#13 of 15
VideoMMMU84.7#7 of 12
V*95.8#3 of 12
OmniDocBench 1.590.8#5 of 11
Video-MME (with subtitle)87.5#6 of 11
MLVU (M-Avg)86.7#3 of 9
CountBench97.2#3 of 8
RefCOCO (avg)92.3#2 of 8
Video-MME (w/o subtitle)83.7#3 of 7
CC-OCR82.0#1 of 7
AI2D_TEST93.9#2 of 7
DynaMath86.3#2 of 6
We-Math87.9#1 of 5
MMLongBench-Doc61.5#2 of 5
ODINW1347.0#3 of 4

Agentic & tool use

Gert Labs46.76#27 of 52
BrowseComp69.0#32 of 50
MCP Atlas46.1#38 of 42
Terminal-Bench 2.052.5#26 of 39
Claw-Eval56.8#28 of 36
OSWorld-Verified62.2#29 of 35
τ³-bench results68.4#8 of 23
Toolathlon36.3#19 of 22
ResearchClawBench14.2#15 of 19
WideResearch74.0#10 of 16
VITA-Bench43.7#4 of 11
QwenClawBench51.8#10 of 10
DeepPlanning37.6#3 of 7
MCP-Tasks74.2#1 of 5

Multilingual

SWE Multilingual69.3#35 of 46
MMLU-ProX84.7#4 of 16
INCLUDE85.6#5 of 10
NOVA-6359.1#1 of 7
PolyMath73.3#4 of 6
MAXIFE88.2#3 of 5

Knowledge

GPQA88.4#32 of 117
HLE28.7#47 of 69
MMLU-Pro87.8#5 of 67
SuperGPQA70.4#9 of 20
MMLU-Redux94.9#3 of 15
C-Eval93.0#3 of 10

Math

AIME2693.3#18 of 29
HMMT Feb 202687.9#9 of 24
HMMT Feb 202594.8#6 of 13
HMMT Nov 202592.7#6 of 12
MMAnswerBench80.9#8 of 10

Coding

SWE-bench Verified76.2#38 of 98
SWE-bench Pro50.9#60 of 69
LiveCodeBench v683.6#21 of 46
NL2Repo32.2#29 of 31

Reasoning

LongBench v263.2#4 of 17
AI-Needle68.7#2 of 4

Instruction following

IFEval92.6#11 of 53
IFBench76.5#19 of 42

Pricing by source

USD per 1M tokens.

SourceInputOutputCached in
List price (BenchLM)$0.60$3.60—

Alibaba Cloud Model Studio's pricing page lists qwen3.5-397b-a17b at $0.60 input / $3.60 output per million tokens in International deployment mode for prompts up to 256K tokens.

Speed

Measured by us through OpenRouter (median of trials).

Not available on OpenRouter, so we cannot measure it yet.

Related models

Similar score