LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

Gemma 4 31B vs Hy3

Updated October 21, 2026. 33 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace Gemma 4 31B Replace Hy3
Gemma 4 31BGoogle
Coding
38.3
Overall
14.7
Output $/M
$0.40
17
Hy3Tencent
Coding
43.9
Overall
No data
Output $/M
$0.528
16
vs boards won of 33 shared

Decision reading

Pick Hy3 for coding: it scores 43.9 on the Vibe Coding Index against 38.3 for Gemma 4 31B, though on 1 benchmark against 2.

Pick Gemma 4 31B when output tokens cost 1.3x less, or when what matters is ALE-Bench or LMArena Text: software and IT, where it leads.

Model AGemma 4 31BGoogle38.3/1002 independent benchmarksNo cost per solved taskModel BHy3Tencent43.9/1001 independent benchmarkNo cost per solved task
Shared results
33
Gemma 4 31B only
35
Hy3 only
11
Like-for-like categories
5 of 9

Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.

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 4 31B 38.3 from 2 benchmarks, Hy3 43.9 from 1 benchmark.
Widest category
Coding, 10.0 points apart on average over 7 shared benchmarks.
Widest single gap
LMArena WebDev: React: Gemma 4 31B 1359 against Hy3 1511, 25.2 points apart on a 0 to 100 scale.
Output price
Gemma 4 31B $0.40 against Hy3 $0.528 per million output tokens, 1.3x apart.
Context window
Both take 262K tokens.

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 4 31BBarsHy3Ties
Coding25
Reasoning32
Math02
Knowledge22
General105
All shared boards1716
CodingLike-for-likeFour of these boards feed the Vibe Coding Index25
Coding benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
LiveBench coding49.0%Not measuredOnly Gemma 4 31B measured
LMArena WebDev13671507Hy3
AA Coding Index43.4Not measuredOnly Gemma 4 31B measured
SciCode45.5%48.6%Hy3
Terminal-Bench Hard (Artificial Analysis run)36.4%Not measuredOnly Gemma 4 31B measured
WeirdML52.3%Not measuredOnly Gemma 4 31B measured
ALE-Bench925.5838.6Gemma 4 31B
LMArena Text: coding14591466Hy3
LMArena Text: software and IT14631463Gemma 4 31B
LMArena WebDev: React13591511Hy3
LMArena WebDev: HTML13801487Hy3
Neither model has a result on the other 27 boards here
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
Intelligence benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
AA Intelligence Index14.7Not measuredOnly Gemma 4 31B measured
Epoch Capabilities Index143.0Not measuredOnly Gemma 4 31B measured
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
Agentic work benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
AA Agentic Index4.2Not measuredOnly Gemma 4 31B measured
LiveBench research44.4%Not measuredOnly Gemma 4 31B measured
LiveBench browser4.0%Not measuredOnly Gemma 4 31B measured
LiveBench long agentic13.3%Not measuredOnly Gemma 4 31B measured
τ²-Bench Telecom65.5%Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 34 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index32
Reasoning benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
LiveBench reasoning68.3%Not measuredOnly Gemma 4 31B measured
GPQA Diamond (Artificial Analysis run)85.7%89.7%Hy3
Humanity's Last Exam (Artificial Analysis run)23.6%33.5%Hy3
GPQA Diamond (Epoch AI run)75.8%Not measuredOnly Gemma 4 31B measured
LMArena Text: hard prompts14481447Gemma 4 31B
LMArena Text: expert14641461Gemma 4 31B
LMArena Text: hard prompts in English14561455Gemma 4 31B
LMArena Vision: diagrams1283Not measuredOnly Gemma 4 31B measured
LMArena Vision: homework1310Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 12 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index02
Math benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
OTIS Mock AIME73.3%Not measuredOnly Gemma 4 31B measured
LMArena Text: math14631473Hy3
LMArena Text: mathematical work14651482Hy3
Neither model has a result on the other 10 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index22
Knowledge benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
CaseLaw (Vals AI)52.6%Not measuredOnly Gemma 4 31B measured
MortgageTax (Vals AI)61.4%Not measuredOnly Gemma 4 31B measured
SimpleQA Verified10.4%Not measuredOnly Gemma 4 31B measured
LMArena Text: science14571457Hy3
LMArena Text: medicine and healthcare14511452Hy3
LMArena Text: legal and government14381422Gemma 4 31B
LMArena Text: business and finance14301424Gemma 4 31B
Neither model has a result on the other 10 boards here
Web designDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
Web design benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
Design Arena: websitesNot measured1230Only Hy3 measured
Design Arena: UI componentsNot measured1205Only Hy3 measured
Design Arena: game devNot measured1208Only Hy3 measured
Design Arena: data vizNot measured1163Only Hy3 measured
Design Arena: 3DNot measured1233Only Hy3 measured
Design Arena: all code categoriesNot measured1227Only Hy3 measured
Design Arena: SVGNot measured1199Only Hy3 measured
Design Arena: ASCII artNot measured1157Only Hy3 measured
LiveBench design58.4%Not measuredOnly Gemma 4 31B measured
GeneralLike-for-likePublic boards, shown as published, kept out of the index105
General benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
LiveBench data analysis53.3%Not measuredOnly Gemma 4 31B measured
SAGE (Vals AI)55.0%Not measuredOnly Gemma 4 31B measured
IFBench75.6%Not measuredOnly Gemma 4 31B measured
Long context reasoning (AA-LCR)69.7%79.0%Hy3
DTBench82.7%Not measuredOnly Gemma 4 31B measured
LMCA39.3%Not measuredOnly Gemma 4 31B measured
Surface Evolver Bench30.6%Not measuredOnly Gemma 4 31B measured
LMArena Text14431438Gemma 4 31B
LMArena Text: instruction following14331426Gemma 4 31B
LMArena Text: creative writing14171403Gemma 4 31B
LMArena Vision1276Not measuredOnly Gemma 4 31B measured
LMArena Text: English14511448Gemma 4 31B
LMArena Text: non-English14301426Gemma 4 31B
LMArena Text: Chinese14751485Hy3
LMArena Text: GermanNot measured1449Only Hy3 measured
LMArena Text: JapaneseNot measured1397Only Hy3 measured
LMArena Text: KoreanNot measured1396Only Hy3 measured
LMArena Text: French14361456Hy3
LMArena Text: Spanish14421457Hy3
LMArena Text: Russian14601432Gemma 4 31B
LMArena Text: Polish14331447Hy3
LMArena Text: multi-turn14521436Gemma 4 31B
LMArena Text: longer queries14461442Gemma 4 31B
LMArena Text: writing and language14321414Gemma 4 31B
LMArena Text: entertainment and media14051400Gemma 4 31B
LMArena Vision: English1271Not measuredOnly Gemma 4 31B measured
LMArena Vision: Chinese1326Not measuredOnly Gemma 4 31B measured
LMArena Vision: reading text1282Not measuredOnly Gemma 4 31B measured
LMArena Vision: humor1280Not measuredOnly Gemma 4 31B measured
LMArena Vision: captioning1218Not measuredOnly Gemma 4 31B measured
LMArena Vision: recognizing things1226Not measuredOnly Gemma 4 31B measured
LMArena Vision: creative writing1285Not measuredOnly Gemma 4 31B measured
LMArena Document1425Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 13 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Gemma 4 31B and Hy3
BenchmarkGemma 4 31BHy3Leader
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 4 31B$0.0004Fits in one request
Hy3$0.0004Fits in one request

Hy3 costs less

Repository review

50K input + 3K output tokens

Gemma 4 31B$0.011Fits in one request
Hy3$0.0082Fits in one request

Hy3 costs less

Agent loop

200K input + 10K output tokens

Gemma 4 31B$0.044Fits in one request
Hy3$0.032Fits in one request

Hy3 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 4 31B and Hy3
SpecificationGoogle Gemma 4 31BTencent Hy3
MakerGoogleTencent
WeightsOpen weightsOpen weights
ReleasedApril 2, 2026July 6, 2026
Context window262K262K
Input $/Mtok$0.20$0.132
Output $/Mtok$0.40$0.528
Coding index38.343.9
Independent measurements behind it21
Cost per solved taskNot computableNot computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Gemma 4 31B better than Hy3 for coding?

Pick Hy3 for coding: it scores 43.9 on the Vibe Coding Index against 38.3 for Gemma 4 31B, though on 1 benchmark against 2. Pick Gemma 4 31B when output tokens cost 1.3x less, or when what matters is ALE-Bench or LMArena Text: software and IT, where it leads.

Which is cheaper, Gemma 4 31B or Hy3?

Gemma 4 31B costs $0.20 per million input tokens and $0.40 per million output tokens; Hy3 costs $0.132 and $0.528. Cost per solved task is known for neither of them.

Which has the bigger context window, Gemma 4 31B or Hy3?

Neither: both take 262K tokens.

How many benchmarks is the Gemma 4 31B vs Hy3 comparison based on?

33 benchmarks have a result for both models as of October 21, 2026. Gemma 4 31B's coding index rests on 2 independent benchmarks and Hy3's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next