Enter opens the highlighted result. Esc closes.

Qwen3 235B A22B Thinking 2507 vs Qwen3.8 27B

Qwen3.8 27B leads Qwen3 235B A22B Thinking 2507 on the Artificial Analysis Intelligence Index, 33.7 to 12.7, and Qwen3.8 27B costs 1.4x less per output token ($2.55 vs $3.50 per million). Updated October 22, 2026; 32 benchmarks have a result for both.

Replace Qwen3 235B A22B Thinking 2507 Replace Qwen3.8 27B
Qwen3 235B A22B Thinking 2507Qwen
Coding
No data
Overall
52.9
Output $/M
$3.50
2
Qwen3.8 27BQwen
Coding
74.2
Overall
59.6
Output $/M
$2.55
30
vs boards won of 32 shared

Qwen3 235B A22B Thinking 2507 or Qwen3.8 27B: which to pick

Only Qwen3.8 27B has a coding score, 74.2 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding.

Pick Qwen3 235B A22B Thinking 2507 for work measured by LMArena Text: medicine and healthcare or LMArena Text: creative writing, where it leads.

Model AQwen3 235B A22B Thinking 2507QwenNot benchmarkedNo coding benchmarkNo cost per solved taskModel BQwen3.8 27BQwen74.2/1004 independent benchmarks$1.26 per solved task
Shared results
32
Qwen3 235B A22B Thinking 2507 only
14
Qwen3.8 27B only
53
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
Qwen3 235B A22B Thinking 2507 not benchmarked, Qwen3.8 27B 74.2 from 4 benchmarks.
Widest category
Agentic work, 44.5 points apart on average over 1 shared benchmark.
Widest single gap
AA Coding Index: Qwen3 235B A22B Thinking 2507 22.1 against Qwen3.8 27B 68.1, 46.0 points apart on a 0 to 100 scale.
Output price
Qwen3 235B A22B Thinking 2507 $3.50 against Qwen3.8 27B $2.55 per million output tokens, 1.4x apart.
Context window
Qwen3 235B A22B Thinking 2507 takes 128K tokens, Qwen3.8 27B 1M.

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
CategoryQwen3 235B A22B Thinking 2507BarsQwen3.8 27BTies
Coding03
Intelligence01
Agentic work01
Reasoning03
Math02
Knowledge13
General117
All shared boards230
CodingLike-for-likeSix of these boards feed the Vibe Coding Index03
Coding benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
LiveBench codingNot measured86.8%Only Qwen3.8 27B measured
LMArena WebDevNot measured1593Only Qwen3.8 27B measured
AA Coding Index22.168.1Qwen3.8 27B
LiveCodeBench (Vals AI)Not measured84.0%Only Qwen3.8 27B measured
SWE-bench Verified (Vals AI)Not measured86.0%Only Qwen3.8 27B measured
Vibe Code Bench (Vals AI)Not measured64.8%Only Qwen3.8 27B measured
Code migration (Vals AI)Not measured14.2%Only Qwen3.8 27B measured
ProgramBench (Vals AI)Not measured0.0%Only Qwen3.8 27B measured
IOI (Vals AI)Not measured39.1%Only Qwen3.8 27B measured
Terminal-Bench 2.1 (Vals AI)Not measured58.4%Only Qwen3.8 27B measured
SkillsBench (Vals AI)Not measured38.1%Only Qwen3.8 27B measured
CyberBench (Vals AI)Not measured67.8%Only Qwen3.8 27B measured
SciCodeNot measured46.6%Only Qwen3.8 27B measured
WeirdML41.0%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
LMArena Text: coding14241483Qwen3.8 27B
LMArena Text: software and IT14311478Qwen3.8 27B
LMArena WebDev: ReactNot measured1602Only Qwen3.8 27B measured
LMArena WebDev: HTMLNot measured1570Only Qwen3.8 27B measured
LMArena WebDev: from an imageNot measured1577Only Qwen3.8 27B measured
Neither model has a result on the other 24 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
Intelligence benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
AA Intelligence Index12.733.7Qwen3.8 27B
Epoch Capabilities IndexNot measured151.7Only Qwen3.8 27B measured
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
AA Agentic Index1.345.8Qwen3.8 27B
LiveBench researchNot measured59.4%Only Qwen3.8 27B measured
LiveBench browserNot measured32.0%Only Qwen3.8 27B measured
LiveBench long agenticNot measured27.9%Only Qwen3.8 27B measured
Finance Agent (Vals AI)Not measured48.5%Only Qwen3.8 27B measured
Tax Agent Bench (Vals AI)Not measured30.9%Only Qwen3.8 27B measured
Legal Research Bench (Vals AI)Not measured36.1%Only Qwen3.8 27B measured
Excel Modeling Benchmark (Vals AI)Not measured59.7%Only Qwen3.8 27B measured
Terminal-Bench-Science (Vals AI)Not measured0.0%Only Qwen3.8 27B measured
APEX-AgentsNot measured47.5%Only Qwen3.8 27B measured
Vending-Bench 2-$11Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Neither model has a result on the other 31 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index03
Reasoning benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
GPQA DiamondNot measured88.9%Only Qwen3.8 27B measured
LiveBench reasoningNot measured80.0%Only Qwen3.8 27B measured
MMMU Pro (Vals AI)Not measured83.9%Only Qwen3.8 27B measured
GPQA Diamond (Artificial Analysis run)Not measured90.5%Only Qwen3.8 27B measured
Humanity's Last Exam (Artificial Analysis run)Not measured33.9%Only Qwen3.8 27B measured
GPQA Diamond (Epoch AI run)80.1%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
ARC-AGINot measured87.5%Only Qwen3.8 27B measured
ARC-AGI-2Not measured42.4%Only Qwen3.8 27B measured
Mystery game puzzles9.0%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
LMArena Text: hard prompts14161459Qwen3.8 27B
LMArena Text: expert14631479Qwen3.8 27B
LMArena Text: hard prompts in English14321461Qwen3.8 27B
LMArena Vision: diagramsNot measured1285Only Qwen3.8 27B measured
LMArena Vision: homeworkNot measured1295Only Qwen3.8 27B measured
Neither model has a result on the other 9 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index02
Math benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
OTIS Mock AIME86.7%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
ProofBenchNot measured16.0%Only Qwen3.8 27B measured
LMArena Text: math14121455Qwen3.8 27B
LMArena Text: mathematical work14351458Qwen3.8 27B
Neither model has a result on the other 10 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index13
Knowledge benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
MMLU ProNot measured84.3%Only Qwen3.8 27B measured
LegalBench (Vals AI)Not measured82.4%Only Qwen3.8 27B measured
TaxEval (Vals AI)Not measured70.8%Only Qwen3.8 27B measured
MortgageTax (Vals AI)Not measured64.9%Only Qwen3.8 27B measured
MedCode (Vals AI)Not measured28.7%Only Qwen3.8 27B measured
SimpleQA Verified40.4%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
LMArena Text: science14331467Qwen3.8 27B
LMArena Text: medicine and healthcare14561437Qwen3 235B A22B Thinking 2507
LMArena Text: legal and government14151447Qwen3.8 27B
LMArena Text: business and finance14131438Qwen3.8 27B
Neither model has a result on the other 10 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
Web design benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
Design Arena: websites1052Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Design Arena: UI components942Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Design Arena: game dev956Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Design Arena: data viz950Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Design Arena: 3D997Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Design Arena: all code categories1032Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
LiveBench designNot measured75.9%Only Qwen3.8 27B measured
Neither model has a result on the other 2 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index117
General benchmarks for Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
LiveBench data analysisNot measured76.6%Only Qwen3.8 27B measured
SAGE (Vals AI)Not measured52.4%Only Qwen3.8 27B measured
MedScribe (Vals AI)Not measured83.8%Only Qwen3.8 27B measured
Long context reasoning (AA-LCR)Not measured82.0%Only Qwen3.8 27B measured
DTBench80.3%88.0%Qwen3.8 27B
LMCA29.3%41.4%Qwen3.8 27B
Fiction.LiveBench75.0%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Creative writing (Lech Mazur)82.4%Not measuredOnly Qwen3 235B A22B Thinking 2507 measured
Surface Evolver BenchNot measured45.0%Only Qwen3.8 27B measured
LMArena Text14151440Qwen3.8 27B
LMArena Text: instruction following13861436Qwen3.8 27B
LMArena Text: creative writing13841383Qwen3 235B A22B Thinking 2507
LMArena VisionNot measured1269Only Qwen3.8 27B measured
LMArena Text: English14291448Qwen3.8 27B
LMArena Text: non-English13971428Qwen3.8 27B
LMArena Text: Chinese14811501Qwen3.8 27B
LMArena Text: German14081443Qwen3.8 27B
LMArena Text: Japanese13861387Qwen3.8 27B
LMArena Text: Korean13751384Qwen3.8 27B
LMArena Text: FrenchNot measured1465Only Qwen3.8 27B measured
LMArena Text: Spanish13891449Qwen3.8 27B
LMArena Text: Russian14011415Qwen3.8 27B
LMArena Text: Polish13901439Qwen3.8 27B
LMArena Text: multi-turn14061440Qwen3.8 27B
LMArena Text: longer queries14001448Qwen3.8 27B
LMArena Text: writing and language13881405Qwen3.8 27B
LMArena Text: entertainment and media13701393Qwen3.8 27B
LMArena Vision: EnglishNot measured1267Only Qwen3.8 27B measured
LMArena Vision: ChineseNot measured1326Only Qwen3.8 27B measured
LMArena Vision: reading textNot measured1281Only Qwen3.8 27B measured
LMArena Vision: humorNot measured1235Only Qwen3.8 27B measured
LMArena Vision: creative writingNot measured1237Only Qwen3.8 27B 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 Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
BenchmarkQwen3 235B A22B Thinking 2507Qwen3.8 27BLeader
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

Qwen3 235B A22B Thinking 2507$0.0022Fits in one request
Qwen3.8 27B$0.0017Fits in one request

Qwen3.8 27B costs less

Repository review

50K input + 3K output tokens

Qwen3 235B A22B Thinking 2507$0.033Fits in one request
Qwen3.8 27B$0.029Fits in one request

Qwen3.8 27B costs less

Agent loop

200K input + 10K output tokens

Qwen3 235B A22B Thinking 2507$0.125Exceeds the context window
Qwen3.8 27B$0.111Fits in one request

Qwen3.8 27B 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 Qwen3 235B A22B Thinking 2507 and Qwen3.8 27B
SpecificationQwen3 235B A22B Thinking 2507Qwen3.8 27B
MakerQwenQwen
WeightsOpen weightsOpen weights
ReleasedJuly 25, 2025August 14, 2026
Context window128K1M
Input $/Mtok$0.45$0.425
Output $/Mtok$3.50$2.55
Coding indexNot benchmarked74.2
Independent measurements behind it04
Cost per solved taskNot computable$1.26
SourceOpenRouter listingOpenRouter listing

Questions

Is Qwen3 235B A22B Thinking 2507 better than Qwen3.8 27B for coding?

Only Qwen3.8 27B has a coding score, 74.2 on the Vibe Coding Index; Qwen3 235B A22B Thinking 2507 has none, so the two cannot be ranked on coding. Pick Qwen3 235B A22B Thinking 2507 for work measured by LMArena Text: medicine and healthcare or LMArena Text: creative writing, where it leads.

Which is cheaper, Qwen3 235B A22B Thinking 2507 or Qwen3.8 27B?

Qwen3 235B A22B Thinking 2507 costs $0.45 per million input tokens and $3.50 per million output tokens; Qwen3.8 27B costs $0.425 and $2.55. Cost per solved task is known for Qwen3.8 27B only.

Which has the bigger context window, Qwen3 235B A22B Thinking 2507 or Qwen3.8 27B?

Qwen3.8 27B, with 1M tokens against 128K.

How many benchmarks is the Qwen3 235B A22B Thinking 2507 vs Qwen3.8 27B comparison based on?

32 benchmarks have a result for both models as of October 22, 2026. Qwen3 235B A22B Thinking 2507's coding index rests on 0 independent benchmarks and Qwen3.8 27B's on 4. Where only one model has a result, the page says so instead of naming a winner.

Compare next