Enter opens the highlighted result. Esc closes.

Gemma 3 4B vs Qwen2.5 72B Instruct

Qwen2.5 72B Instruct leads Gemma 3 4B on the Artificial Analysis Intelligence Index, 7.7 to 4.8, and Gemma 3 4B costs 5.3x less per output token ($0.075 vs $0.40 per million). Updated October 21, 2026; 37 benchmarks have a result for both.

Replace Gemma 3 4B Replace Qwen2.5 72B Instruct
Gemma 3 4BGoogle
Coding
No data
Overall
22.6
Output $/M
$0.075
13
Qwen2.5 72B InstructQwen
Coding
No data
Overall
26.5
Output $/M
$0.40
24
vs boards won of 37 shared

Gemma 3 4B or Qwen2.5 72B Instruct: which to pick

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

Pick Gemma 3 4B for output tokens 5.3x cheaper ($0.075 vs $0.40 per million) and a larger context window (131K against 33K tokens).

Model AGemma 3 4BGoogleNot benchmarkedNo coding benchmarkNo cost per solved taskModel BQwen2.5 72B InstructQwenNot benchmarkedNo coding benchmarkNo cost per solved task
Shared results
37
Gemma 3 4B only
5
Qwen2.5 72B Instruct only
16
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
Gemma 3 4B not benchmarked, Qwen2.5 72B Instruct not benchmarked.
Widest category
Agentic work, 29.5 points apart on average over 1 shared benchmark.
Widest single gap
MMLU Pro (Artificial Analysis run): Gemma 3 4B 41.7% against Qwen2.5 72B Instruct 72.0%, 30.3 points apart on a 0 to 100 scale.
Output price
Gemma 3 4B $0.075 against Qwen2.5 72B Instruct $0.40 per million output tokens, 5.3x apart.
Context window
Gemma 3 4B takes 131K tokens, Qwen2.5 72B Instruct 33K.

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
CategoryGemma 3 4BBarsQwen2.5 72B InstructTies
Coding04
Intelligence02
Agentic work01
Reasoning15
Math05
Knowledge41
General86
All shared boards1324
CodingLike-for-likeSix of these boards feed the Vibe Coding Index04
Coding benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
AA Coding Index2.7Not measuredOnly Gemma 3 4B measured
LiveCodeBench (Artificial Analysis run)11.2%27.6%Qwen2.5 72B Instruct
Terminal-Bench Hard (Artificial Analysis run)0.8%4.5%Qwen2.5 72B Instruct
WeirdMLNot measured16.0%Only Qwen2.5 72B Instruct measured
LMArena Text: coding12301292Qwen2.5 72B Instruct
LMArena Text: software and IT12691292Qwen2.5 72B Instruct
SciCode0.0%Not measuredOnly Gemma 3 4B measured
Neither model has a result on the other 32 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
AA Intelligence Index4.87.7Qwen2.5 72B Instruct
Epoch Capabilities Index115.9129.0Qwen2.5 72B Instruct
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
τ²-Bench Telecom5.0%34.5%Qwen2.5 72B Instruct
METR time horizonsNot measured35.8%Only Qwen2.5 72B Instruct measured
BALROGNot measured16.2%Only Qwen2.5 72B Instruct measured
TheAgentCompanyNot measured5.7%Only Qwen2.5 72B Instruct measured
Berkeley Function Calling Leaderboard (BFCL)19.6%Not measuredOnly Gemma 3 4B measured
Neither model has a result on the other 36 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index15
Reasoning benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
GPQA Diamond (Artificial Analysis run)29.1%49.1%Qwen2.5 72B Instruct
Humanity's Last Exam (Artificial Analysis run)5.3%3.6%Gemma 3 4B
GPQA Diamond (Epoch AI run)23.2%49.1%Qwen2.5 72B Instruct
BIG-Bench HardNot measured79.8%Only Qwen2.5 72B Instruct measured
LMArena Text: hard prompts12531271Qwen2.5 72B Instruct
LMArena Text: expert12231245Qwen2.5 72B Instruct
LMArena Text: hard prompts in English12531279Qwen2.5 72B Instruct
Kagi LLM Benchmark25.2%Not measuredOnly Gemma 3 4B measured
Neither model has a result on the other 14 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index05
Math benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
AA Math Index12.714.0Qwen2.5 72B Instruct
AIME 2025 (Artificial Analysis run)12.7%14.0%Qwen2.5 72B Instruct
OTIS Mock AIME7.5%8.1%Qwen2.5 72B Instruct
MATH level 5Not measured63.2%Only Qwen2.5 72B Instruct measured
LMArena Text: math12391283Qwen2.5 72B Instruct
LMArena Text: mathematical work12561290Qwen2.5 72B Instruct
Neither model has a result on the other 7 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index41
Knowledge benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
MMLU Pro (Artificial Analysis run)41.7%72.0%Qwen2.5 72B Instruct
MMLUNot measured85.3%Only Qwen2.5 72B Instruct measured
AI2 Reasoning ChallengeNot measured94.5%Only Qwen2.5 72B Instruct measured
LMArena Text: science13061275Gemma 3 4B
LMArena Text: medicine and healthcare13421258Gemma 3 4B
LMArena Text: legal and government13271288Gemma 3 4B
LMArena Text: business and finance13111253Gemma 3 4B
Neither model has a result on the other 11 boards here
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Web design benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
Neither model has a result on 9 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index86
General benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
IFBench28.3%36.9%Qwen2.5 72B Instruct
Long context reasoning (AA-LCR)6.7%Not measuredOnly Gemma 3 4B measured
DTBench50.9%62.9%Qwen2.5 72B Instruct
LMCA2.8%13.4%Qwen2.5 72B Instruct
HellaSwagNot measured84.8%Only Qwen2.5 72B Instruct measured
WinoGrandeNot measured82.3%Only Qwen2.5 72B Instruct measured
PIQANot measured82.6%Only Qwen2.5 72B Instruct measured
LMArena Text12911269Gemma 3 4B
LMArena Text: instruction following12391254Qwen2.5 72B Instruct
LMArena Text: creative writing12701221Gemma 3 4B
LMArena Text: English13001283Gemma 3 4B
LMArena Text: non-English12731252Gemma 3 4B
LMArena Text: ChineseNot measured1273Only Qwen2.5 72B Instruct measured
LMArena Text: German12811234Gemma 3 4B
LMArena Text: JapaneseNot measured1180Only Qwen2.5 72B Instruct measured
LMArena Text: KoreanNot measured1188Only Qwen2.5 72B Instruct measured
LMArena Text: FrenchNot measured1280Only Qwen2.5 72B Instruct measured
LMArena Text: SpanishNot measured1255Only Qwen2.5 72B Instruct measured
LMArena Text: Russian12941264Gemma 3 4B
LMArena Text: multi-turn12551272Qwen2.5 72B Instruct
LMArena Text: longer queries12731281Qwen2.5 72B Instruct
LMArena Text: writing and language12881247Gemma 3 4B
LMArena Text: entertainment and media12331211Gemma 3 4B
Neither model has a result on the other 23 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Gemma 3 4B and Qwen2.5 72B Instruct
BenchmarkGemma 3 4BQwen2.5 72B InstructLeader
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

Gemma 3 4B$0.0001Fits in one request
Qwen2.5 72B Instruct$0.0006Fits in one request

Gemma 3 4B costs less

Repository review

50K input + 3K output tokens

Gemma 3 4B$0.0025Fits in one request
Qwen2.5 72B Instruct$0.019Exceeds the context window

Gemma 3 4B costs less

Agent loop

200K input + 10K output tokens

Gemma 3 4B$0.0097Exceeds the context window
Qwen2.5 72B Instruct$0.076Exceeds the context window

Gemma 3 4B 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 Gemma 3 4B and Qwen2.5 72B Instruct
SpecificationGoogle Gemma 3 4BQwen2.5 72B Instruct
MakerGoogleQwen
WeightsOpen weightsOpen weights
ReleasedMarch 13, 2025September 19, 2024
Context window131K33K
Input $/Mtok$0.045$0.36
Output $/Mtok$0.075$0.40
Coding indexNot benchmarkedNot benchmarked
Independent measurements behind it00
Cost per solved taskNot computableNot computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Gemma 3 4B better than Qwen2.5 72B Instruct for coding?

Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Gemma 3 4B for output tokens 5.3x cheaper ($0.075 vs $0.40 per million) and a larger context window (131K against 33K tokens).

Which is cheaper, Gemma 3 4B or Qwen2.5 72B Instruct?

Gemma 3 4B costs $0.045 per million input tokens and $0.075 per million output tokens; Qwen2.5 72B Instruct costs $0.36 and $0.40. Cost per solved task is known for neither of them.

Which has the bigger context window, Gemma 3 4B or Qwen2.5 72B Instruct?

Gemma 3 4B, with 131K tokens against 33K.

How many benchmarks is the Gemma 3 4B vs Qwen2.5 72B Instruct comparison based on?

37 benchmarks have a result for both models as of October 21, 2026. Gemma 3 4B's coding index rests on 0 independent benchmarks and Qwen2.5 72B Instruct's on 0. Where only one model has a result, the page says so instead of naming a winner.

Compare next