Enter opens the highlighted result. Esc closes.

Qwen2.5 72B Instruct vs Qwen3.5 397B A17B

Qwen3.5 397B A17B leads Qwen2.5 72B Instruct on the Artificial Analysis Intelligence Index, 21.4 to 7.7, and Qwen2.5 72B Instruct costs 5.8x less per output token ($0.40 vs $2.34 per million). Updated October 22, 2026; 38 benchmarks have a result for both.

Replace Qwen2.5 72B Instruct Replace Qwen3.5 397B A17B
Qwen2.5 72B InstructQwen
Coding
No data
Overall
26.5
Output $/M
$0.40
0
Qwen3.5 397B A17BQwen
Coding
No data
Overall
57.1
Output $/M
$2.34
38
vs boards won of 38 shared

Qwen2.5 72B Instruct or Qwen3.5 397B A17B: which to pick

Neither model has a coding score yet, so the two cannot be ranked on coding.

Pick Qwen2.5 72B Instruct for output tokens 5.8x cheaper ($0.40 vs $2.34 per million) and work measured by Terminal-Bench Hard (Artificial Analysis run) or LMArena Text: coding, where it leads.

Model AQwen2.5 72B InstructQwenNot benchmarkedNo coding benchmarkNo cost per solved taskModel BQwen3.5 397B A17BQwenNot benchmarked1 independent benchmarkNo cost per solved task
Shared results
38
Qwen2.5 72B Instruct only
15
Qwen3.5 397B A17B only
28
Like-for-like categories
7 of 9

Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.

Which one for your work

Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.

At a glance

Coding index
Qwen2.5 72B Instruct not benchmarked, Qwen3.5 397B A17B not benchmarked.
Widest category
Agentic work, 61.1 points apart on average over 1 shared benchmark.
Widest single gap
OTIS Mock AIME: Qwen2.5 72B Instruct 8.1% against Qwen3.5 397B A17B 88.9%, 80.8 points apart on a 0 to 100 scale.
Output price
Qwen2.5 72B Instruct $0.40 against Qwen3.5 397B A17B $2.34 per million output tokens, 5.8x apart.
Context window
Qwen2.5 72B Instruct takes 33K tokens, Qwen3.5 397B A17B 262K.

Benchmark by benchmark

First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.

Boards won, where both have a result
CategoryQwen2.5 72B InstructBarsQwen3.5 397B A17BTies
Coding03
Intelligence02
Agentic work01
Reasoning06
Math03
Knowledge04
General019
All shared boards038
CodingLike-for-likeSix of these boards feed the Vibe Coding Index03
Coding benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
LMArena WebDevNot measured1400Only Qwen3.5 397B A17B measured
LiveCodeBench (Artificial Analysis run)27.6%Not measuredOnly Qwen2.5 72B Instruct measured
SciCodeNot measured44.8%Only Qwen3.5 397B A17B measured
Terminal-Bench Hard (Artificial Analysis run)4.5%40.9%Qwen3.5 397B A17B
WeirdML16.0%Not measuredOnly Qwen2.5 72B Instruct measured
ALE-BenchNot measured621.9Only Qwen3.5 397B A17B measured
LMArena Text: coding12921464Qwen3.5 397B A17B
LMArena Text: software and IT12921463Qwen3.5 397B A17B
LMArena WebDev: ReactNot measured1388Only Qwen3.5 397B A17B measured
LMArena WebDev: HTMLNot measured1410Only Qwen3.5 397B A17B measured
Neither model has a result on the other 33 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
AA Intelligence Index7.721.4Qwen3.5 397B A17B
Epoch Capabilities Index129.0146.8Qwen3.5 397B A17B
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
τ²-Bench Telecom34.5%95.6%Qwen3.5 397B A17B
APEX-AgentsNot measured24.9%Only Qwen3.5 397B A17B measured
METR time horizons35.8%Not measuredOnly Qwen2.5 72B Instruct measured
BALROG16.2%Not measuredOnly Qwen2.5 72B Instruct measured
TheAgentCompany5.7%Not measuredOnly Qwen2.5 72B Instruct measured
Neither model has a result on the other 37 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index06
Reasoning benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
GPQA Diamond (Artificial Analysis run)49.1%89.3%Qwen3.5 397B A17B
Humanity's Last Exam (Artificial Analysis run)3.6%29.0%Qwen3.5 397B A17B
GPQA Diamond (Epoch AI run)49.1%86.4%Qwen3.5 397B A17B
Mystery game puzzlesNot measured18.0%Only Qwen3.5 397B A17B measured
BIG-Bench Hard79.8%Not measuredOnly Qwen2.5 72B Instruct measured
LMArena Text: hard prompts12711448Qwen3.5 397B A17B
LMArena Text: expert12451462Qwen3.5 397B A17B
LMArena Text: hard prompts in English12791450Qwen3.5 397B A17B
LMArena Vision: diagramsNot measured1277Only Qwen3.5 397B A17B measured
LMArena Vision: homeworkNot measured1292Only Qwen3.5 397B A17B measured
Neither model has a result on the other 13 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index03
Math benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
AA Math Index14.0Not measuredOnly Qwen2.5 72B Instruct measured
AIME 2025 (Artificial Analysis run)14.0%Not measuredOnly Qwen2.5 72B Instruct measured
OTIS Mock AIME8.1%88.9%Qwen3.5 397B A17B
FrontierMath, tiers 1 to 3Not measured31.2%Only Qwen3.5 397B A17B measured
MATH level 563.2%Not measuredOnly Qwen2.5 72B Instruct measured
LMArena Text: math12831453Qwen3.5 397B A17B
LMArena Text: mathematical work12901454Qwen3.5 397B A17B
MathArenaNot measured19.2%Only Qwen3.5 397B A17B measured
Neither model has a result on the other 6 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
Knowledge benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
MMLU Pro (Artificial Analysis run)72.0%Not measuredOnly Qwen2.5 72B Instruct measured
MMLU85.3%Not measuredOnly Qwen2.5 72B Instruct measured
AI2 Reasoning Challenge94.5%Not measuredOnly Qwen2.5 72B Instruct measured
LMArena Text: science12751457Qwen3.5 397B A17B
LMArena Text: medicine and healthcare12581455Qwen3.5 397B A17B
LMArena Text: legal and government12881441Qwen3.5 397B A17B
LMArena Text: business and finance12531435Qwen3.5 397B A17B
Neither model has a result on the other 13 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
Web design benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
Design Arena: websitesNot measured1190Only Qwen3.5 397B A17B measured
Design Arena: UI componentsNot measured1161Only Qwen3.5 397B A17B measured
Design Arena: game devNot measured1140Only Qwen3.5 397B A17B measured
Design Arena: data vizNot measured1175Only Qwen3.5 397B A17B measured
Design Arena: 3DNot measured1160Only Qwen3.5 397B A17B measured
Design Arena: all code categoriesNot measured1179Only Qwen3.5 397B A17B measured
Design Arena: SVGNot measured1139Only Qwen3.5 397B A17B measured
Neither model has a result on the other 2 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index019
General benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
IFBench36.9%78.8%Qwen3.5 397B A17B
Long context reasoning (AA-LCR)Not measured77.3%Only Qwen3.5 397B A17B measured
DTBench62.9%87.5%Qwen3.5 397B A17B
LMCA13.4%37.9%Qwen3.5 397B A17B
HellaSwag84.8%Not measuredOnly Qwen2.5 72B Instruct measured
WinoGrande82.3%Not measuredOnly Qwen2.5 72B Instruct measured
PIQA82.6%Not measuredOnly Qwen2.5 72B Instruct measured
LMArena Text12691437Qwen3.5 397B A17B
LMArena Text: instruction following12541424Qwen3.5 397B A17B
LMArena Text: creative writing12211402Qwen3.5 397B A17B
LMArena VisionNot measured1262Only Qwen3.5 397B A17B measured
LMArena Text: English12831443Qwen3.5 397B A17B
LMArena Text: non-English12521430Qwen3.5 397B A17B
LMArena Text: Chinese12731498Qwen3.5 397B A17B
LMArena Text: German12341445Qwen3.5 397B A17B
LMArena Text: Japanese11801429Qwen3.5 397B A17B
LMArena Text: Korean11881384Qwen3.5 397B A17B
LMArena Text: French12801462Qwen3.5 397B A17B
LMArena Text: Spanish12551441Qwen3.5 397B A17B
LMArena Text: Russian12641430Qwen3.5 397B A17B
LMArena Text: PolishNot measured1447Only Qwen3.5 397B A17B measured
LMArena Text: multi-turn12721446Qwen3.5 397B A17B
LMArena Text: longer queries12811441Qwen3.5 397B A17B
LMArena Text: writing and language12471415Qwen3.5 397B A17B
LMArena Text: entertainment and media12111399Qwen3.5 397B A17B
LMArena Vision: EnglishNot measured1268Only Qwen3.5 397B A17B measured
LMArena Vision: ChineseNot measured1314Only Qwen3.5 397B A17B measured
LMArena Vision: reading textNot measured1271Only Qwen3.5 397B A17B measured
LMArena Vision: humorNot measured1262Only Qwen3.5 397B A17B measured
LMArena Vision: captioningNot measured1230Only Qwen3.5 397B A17B measured
LMArena Vision: recognizing thingsNot measured1265Only Qwen3.5 397B A17B measured
LMArena Vision: creative writingNot measured1260Only Qwen3.5 397B A17B measured
Neither model has a result on the other 16 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Qwen2.5 72B Instruct and Qwen3.5 397B A17B
BenchmarkQwen2.5 72B InstructQwen3.5 397B A17BLeader
Neither model has a result on 5 boards here

What each workload costs

Three fixed token mixes at each model's listed price, so per-token rates become a bill.

Chat turn

1K input + 500 output tokens

Qwen2.5 72B Instruct$0.0006Fits in one request
Qwen3.5 397B A17B$0.0016Fits in one request

Qwen2.5 72B Instruct costs less

Repository review

50K input + 3K output tokens

Qwen2.5 72B Instruct$0.019Exceeds the context window
Qwen3.5 397B A17B$0.027Fits in one request

Qwen2.5 72B Instruct costs less

Agent loop

200K input + 10K output tokens

Qwen2.5 72B Instruct$0.076Exceeds the context window
Qwen3.5 397B A17B$0.101Fits in one request

Qwen2.5 72B Instruct costs less

Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.

Specification differences

The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.

Specifications of Qwen2.5 72B Instruct and Qwen3.5 397B A17B
SpecificationQwen2.5 72B InstructQwen3.5 397B A17B
MakerQwenQwen
WeightsOpen weightsOpen weights
ReleasedSeptember 19, 2024February 16, 2026
Context window33K262K
Input $/Mtok$0.36$0.39
Output $/Mtok$0.40$2.34
Coding indexNot benchmarkedNot benchmarked
Independent measurements behind it01
Cost per solved taskNot computableNot computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Qwen2.5 72B Instruct better than Qwen3.5 397B A17B for coding?

Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Qwen2.5 72B Instruct for output tokens 5.8x cheaper ($0.40 vs $2.34 per million) and work measured by Terminal-Bench Hard (Artificial Analysis run) or LMArena Text: coding, where it leads.

Which is cheaper, Qwen2.5 72B Instruct or Qwen3.5 397B A17B?

Qwen2.5 72B Instruct costs $0.36 per million input tokens and $0.40 per million output tokens; Qwen3.5 397B A17B costs $0.39 and $2.34. Cost per solved task is known for neither of them.

Which has the bigger context window, Qwen2.5 72B Instruct or Qwen3.5 397B A17B?

Qwen3.5 397B A17B, with 262K tokens against 33K.

How many benchmarks is the Qwen2.5 72B Instruct vs Qwen3.5 397B A17B comparison based on?

38 benchmarks have a result for both models as of October 22, 2026. Qwen2.5 72B Instruct's coding index rests on 0 independent benchmarks and Qwen3.5 397B A17B's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next