Enter opens the highlighted result. Esc closes.

GPT-5.4 vs Qwen3 235B A22B Thinking 2507

GPT-5.4 leads Qwen3 235B A22B Thinking 2507 on the Artificial Analysis Intelligence Index, 39.0 to 12.7, and Qwen3 235B A22B Thinking 2507 costs 4.3x less per output token ($3.50 vs $15.00 per million). Updated October 22, 2026; 43 benchmarks have a result for both.

Replace GPT-5.4 Replace Qwen3 235B A22B Thinking 2507
GPT-5.4OpenAI
Coding
64.4
Overall
69.6
Output $/M
$15.00
43
Qwen3 235B A22B Thinking 2507Qwen
Coding
No data
Overall
52.9
Output $/M
$3.50
0
vs boards won of 43 shared

GPT-5.4 or Qwen3 235B A22B Thinking 2507: which to pick

Only GPT-5.4 has a coding score, 64.4 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 output tokens 4.3x cheaper ($3.50 vs $15.00 per million).

Model AGPT-5.4OpenAI64.4/1003 independent benchmarks$1.02 per solved taskModel BQwen3 235B A22B Thinking 2507QwenNot benchmarkedNo coding benchmarkNo cost per solved task
Shared results
43
GPT-5.4 only
75
Qwen3 235B A22B Thinking 2507 only
3
Like-for-like categories
8 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
GPT-5.4 64.4 from 3 benchmarks, Qwen3 235B A22B Thinking 2507 not benchmarked.
Widest category
Agentic work, 38.4 points apart on average over 1 shared benchmark.
Widest single gap
AA Coding Index: GPT-5.4 71.1 against Qwen3 235B A22B Thinking 2507 22.1, 49.0 points apart on a 0 to 100 scale.
Output price
GPT-5.4 $15.00 against Qwen3 235B A22B Thinking 2507 $3.50 per million output tokens, 4.3x apart.
Context window
GPT-5.4 takes 1.05M tokens, Qwen3 235B A22B Thinking 2507 128K.

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
CategoryGPT-5.4BarsQwen3 235B A22B Thinking 2507Ties
Coding40
Intelligence10
Agentic work10
Reasoning50
Math30
Knowledge50
Web design60
General180
All shared boards430
CodingLike-for-likeSix of these boards feed the Vibe Coding Index40
Coding benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
LMArena WebDev1464Not measuredOnly GPT-5.4 measured
AA Coding Index71.122.1GPT-5.4
LiveCodeBench (Vals AI)84.1%Not measuredOnly GPT-5.4 measured
SWE-bench Verified (Vals AI)78.2%Not measuredOnly GPT-5.4 measured
Vibe Code Bench (Vals AI)67.4%Not measuredOnly GPT-5.4 measured
Code migration (Vals AI)35.0%Not measuredOnly GPT-5.4 measured
ProgramBench (Vals AI)0.5%Not measuredOnly GPT-5.4 measured
SkillsBench (Vals AI)51.7%Not measuredOnly GPT-5.4 measured
Terminal-Bench Hard (Artificial Analysis run)57.6%Not measuredOnly GPT-5.4 measured
Terminal-Bench81.8%Not measuredOnly GPT-5.4 measured
GSO-Bench31.4%Not measuredOnly GPT-5.4 measured
WeirdML77.7%41.0%GPT-5.4
MirrorCode15.6%Not measuredOnly GPT-5.4 measured
ALE-Bench1607.0Not measuredOnly GPT-5.4 measured
LMArena Text: coding14961424GPT-5.4
LMArena Text: software and IT14911431GPT-5.4
LMArena WebDev: React1454Not measuredOnly GPT-5.4 measured
LMArena WebDev: HTML1478Not measuredOnly GPT-5.4 measured
LMArena WebDev: from an image1436Not measuredOnly GPT-5.4 measured
SWE Atlas: Codebase QnA40.8%Not measuredOnly GPT-5.4 measured
SWE Atlas: Refactoring44.3%Not measuredOnly GPT-5.4 measured
SWE Atlas: Test Writing44.4%Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 21 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
Intelligence benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
AA Intelligence Index39.012.7GPT-5.4
Epoch Capabilities Index157.3Not measuredOnly GPT-5.4 measured
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
Agentic work benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
AA Agentic IndexNot measured1.3Only Qwen3 235B A22B Thinking 2507 measured
Design Arena: agents building web apps1059Not measuredOnly GPT-5.4 measured
Design Arena: agents building full-stack apps1017Not measuredOnly GPT-5.4 measured
Design Arena: agents building mobile apps1086Not measuredOnly GPT-5.4 measured
Design Arena: agents building Android apps1031Not measuredOnly GPT-5.4 measured
Design Arena: agents building Godot games1135Not measuredOnly GPT-5.4 measured
τ²-Bench Telecom87.1%Not measuredOnly GPT-5.4 measured
APEX-Agents52.4%Not measuredOnly GPT-5.4 measured
METR time horizons74.3%Not measuredOnly GPT-5.4 measured
DeepResearch Bench35.1%Not measuredOnly GPT-5.4 measured
PostTrainBench19.0%Not measuredOnly GPT-5.4 measured
Vending-Bench 2$6,144-$11GPT-5.4
MCP Atlas70.6%Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 29 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index50
Reasoning benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
GPQA Diamond91.7%Not measuredOnly GPT-5.4 measured
MMMU Pro (Vals AI)87.5%Not measuredOnly GPT-5.4 measured
GPQA Diamond (Artificial Analysis run)92.0%Not measuredOnly GPT-5.4 measured
Humanity's Last Exam (Artificial Analysis run)43.7%Not measuredOnly GPT-5.4 measured
GPQA Diamond (Epoch AI run)93.3%80.1%GPT-5.4
ARC-AGI93.7%Not measuredOnly GPT-5.4 measured
ARC-AGI-274.0%Not measuredOnly GPT-5.4 measured
Chess puzzles48.0%Not measuredOnly GPT-5.4 measured
Mystery game puzzles37.0%9.0%GPT-5.4
EBR-bench25.4%Not measuredOnly GPT-5.4 measured
LMArena Text: hard prompts14851416GPT-5.4
LMArena Text: expert15071463GPT-5.4
LMArena Text: hard prompts in English14791432GPT-5.4
LMArena Vision: diagrams1327Not measuredOnly GPT-5.4 measured
LMArena Vision: homework1336Not measuredOnly GPT-5.4 measured
Kagi LLM Benchmark63.8%Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 7 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index30
Math benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
AIME96.7%Not measuredOnly GPT-5.4 measured
OTIS Mock AIME97.8%86.7%GPT-5.4
FrontierMath, tiers 1 to 378.6%Not measuredOnly GPT-5.4 measured
FrontierMath, tier 449.0%Not measuredOnly GPT-5.4 measured
ProofBench56.0%Not measuredOnly GPT-5.4 measured
LMArena Text: math14881412GPT-5.4
LMArena Text: mathematical work14971435GPT-5.4
MathArena37.2%Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 6 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index50
Knowledge benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
MMLU Pro87.5%Not measuredOnly GPT-5.4 measured
LegalBench (Vals AI)86.0%Not measuredOnly GPT-5.4 measured
CaseLaw (Vals AI)63.8%Not measuredOnly GPT-5.4 measured
CorpFin (Vals AI)65.3%Not measuredOnly GPT-5.4 measured
TaxEval (Vals AI)74.0%Not measuredOnly GPT-5.4 measured
MortgageTax (Vals AI)68.3%Not measuredOnly GPT-5.4 measured
MedQA (Vals AI)96.1%Not measuredOnly GPT-5.4 measured
MedCode (Vals AI)41.3%Not measuredOnly GPT-5.4 measured
SimpleQA Verified45.1%40.4%GPT-5.4
LMArena Text: science14791433GPT-5.4
LMArena Text: medicine and healthcare14581456GPT-5.4
LMArena Text: legal and government14861415GPT-5.4
LMArena Text: business and finance14741413GPT-5.4
PRBench Finance45.6%Not measuredOnly GPT-5.4 measured
PRBench Legal44.4%Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 5 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index60
Web design benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
Design Arena: websites11981052GPT-5.4
Design Arena: UI components1213942GPT-5.4
Design Arena: game dev1194956GPT-5.4
Design Arena: data viz1196950GPT-5.4
Design Arena: 3D1065997GPT-5.4
Design Arena: all code categories11791032GPT-5.4
Design Arena: SVG1141Not measuredOnly GPT-5.4 measured
Design Arena: ASCII art1188Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 1 board here
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
General benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
SAGE (Vals AI)43.3%Not measuredOnly GPT-5.4 measured
MedScribe (Vals AI)77.5%Not measuredOnly GPT-5.4 measured
IFBench73.9%Not measuredOnly GPT-5.4 measured
Long context reasoning (AA-LCR)82.0%Not measuredOnly GPT-5.4 measured
DTBench94.4%80.3%GPT-5.4
LMCA52.0%29.3%GPT-5.4
Fiction.LiveBenchNot measured75.0%Only Qwen3 235B A22B Thinking 2507 measured
Creative writing (Lech Mazur)Not measured82.4%Only Qwen3 235B A22B Thinking 2507 measured
CL-bench27.9%Not measuredOnly GPT-5.4 measured
Furniture assembly37.5%Not measuredOnly GPT-5.4 measured
CL-bench Life21.7%Not measuredOnly GPT-5.4 measured
LMArena Text14691415GPT-5.4
LMArena Text: instruction following14691386GPT-5.4
LMArena Text: creative writing14391384GPT-5.4
LMArena Vision1303Not measuredOnly GPT-5.4 measured
LMArena Text: English14671429GPT-5.4
LMArena Text: non-English14651397GPT-5.4
LMArena Text: Chinese15201481GPT-5.4
LMArena Text: German14721408GPT-5.4
LMArena Text: Japanese14851386GPT-5.4
LMArena Text: Korean14491375GPT-5.4
LMArena Text: French1493Not measuredOnly GPT-5.4 measured
LMArena Text: Spanish14521389GPT-5.4
LMArena Text: Russian14801401GPT-5.4
LMArena Text: Polish14721390GPT-5.4
LMArena Text: multi-turn14821406GPT-5.4
LMArena Text: longer queries14731400GPT-5.4
LMArena Text: writing and language14581388GPT-5.4
LMArena Text: entertainment and media14381370GPT-5.4
LMArena Vision: English1301Not measuredOnly GPT-5.4 measured
LMArena Vision: Chinese1354Not measuredOnly GPT-5.4 measured
LMArena Vision: reading text1312Not measuredOnly GPT-5.4 measured
LMArena Vision: humor1284Not measuredOnly GPT-5.4 measured
LMArena Vision: recognizing things1251Not measuredOnly GPT-5.4 measured
LMArena Vision: creative writing1283Not measuredOnly GPT-5.4 measured
LMArena Document1471Not measuredOnly GPT-5.4 measured
EQ-Bench 41272Not measuredOnly GPT-5.4 measured
Neither model has a result on the other 11 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for GPT-5.4 and Qwen3 235B A22B Thinking 2507
BenchmarkGPT-5.4Qwen3 235B A22B Thinking 2507Leader
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

GPT-5.4$0.010Fits in one request
Qwen3 235B A22B Thinking 2507$0.0022Fits in one request

Qwen3 235B A22B Thinking 2507 costs less

Repository review

50K input + 3K output tokens

GPT-5.4$0.170Fits in one request
Qwen3 235B A22B Thinking 2507$0.033Fits in one request

Qwen3 235B A22B Thinking 2507 costs less

Agent loop

200K input + 10K output tokens

GPT-5.4$0.650Fits in one request
Qwen3 235B A22B Thinking 2507$0.125Exceeds the context window

Qwen3 235B A22B Thinking 2507 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 GPT-5.4 and Qwen3 235B A22B Thinking 2507
SpecificationOpenAI GPT-5.4Qwen3 235B A22B Thinking 2507
MakerOpenAIQwen
WeightsNo public weights listedOpen weights
ReleasedMarch 5, 2026July 25, 2025
Context window1.05M128K
Input $/Mtok$2.50$0.45
Output $/Mtok$15.00$3.50
Coding index64.4Not benchmarked
Independent measurements behind it30
Cost per solved task$1.02Not computable
SourceOpenRouter listingOpenRouter listing

Questions

Is GPT-5.4 better than Qwen3 235B A22B Thinking 2507 for coding?

Only GPT-5.4 has a coding score, 64.4 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 output tokens 4.3x cheaper ($3.50 vs $15.00 per million).

Which is cheaper, GPT-5.4 or Qwen3 235B A22B Thinking 2507?

GPT-5.4 costs $2.50 per million input tokens and $15.00 per million output tokens; Qwen3 235B A22B Thinking 2507 costs $0.45 and $3.50. Cost per solved task is known for GPT-5.4 only.

Which has the bigger context window, GPT-5.4 or Qwen3 235B A22B Thinking 2507?

GPT-5.4, with 1.05M tokens against 128K.

How many benchmarks is the GPT-5.4 vs Qwen3 235B A22B Thinking 2507 comparison based on?

43 benchmarks have a result for both models as of October 22, 2026. GPT-5.4's coding index rests on 3 independent benchmarks and Qwen3 235B A22B Thinking 2507's on 0. Where only one model has a result, the page says so instead of naming a winner.

Compare next