Enter opens the highlighted result. Esc closes.

Gemma 4 31B vs GPT-6.2 Astra

GPT-6.2 Astra leads Gemma 4 31B on coding, 87.0 to 21.9 on the Vibe Coding Index, and Gemma 4 31B costs 100.0x less per output token ($0.40 vs $40.00 per million). Updated November 22, 2026; 38 benchmarks have a result for both.

Gemma 4 31BGoogle
Coding
21.9
Overall
35.6
Output $/M
$0.40
14
GPT-6.2 AstraOpenAI
Coding
87.0
Overall
86.4
Output $/M
$40.00
24
vs boards won of 38 shared

Gemma 4 31B or GPT-6.2 Astra: which to pick

Pick GPT-6.2 Astra for coding: it scores 87.0 on the Vibe Coding Index against 21.9 for Gemma 4 31B.

Pick Gemma 4 31B for output tokens 100.0x cheaper ($0.40 vs $40.00 per million) and work measured by LMArena Text: coding or LMArena Text: software and IT, where it leads.

Shared results
38
Gemma 4 31B only
39
GPT-6.2 Astra only
38
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
Gemma 4 31B 21.9 from 2 benchmarks, GPT-6.2 Astra 87.0 from 4 benchmarks.
Widest category
Agentic work, 69.6 points apart on average over 3 shared benchmarks.
Widest single gap
LiveBench browser: Gemma 4 31B 0.0% against GPT-6.2 Astra 93.0%, 93.0 points apart on a 0 to 100 scale.
Output price
Gemma 4 31B $0.40 against GPT-6.2 Astra $40.00 per million output tokens, 100.0x apart.
Context window
Gemma 4 31B takes 262K tokens, GPT-6.2 Astra 1.05M.

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 31BBarsGPT-6.2 AstraTies
Coding27
Intelligence02
Agentic work03
Reasoning24
Math01
Knowledge32
Web design01
General74
All shared boards1424
CodingLike-for-likeSix of these boards feed the Vibe Coding Index27
Coding benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
LiveBench coding20.3%91.9%GPT-6.2 Astra
LMArena WebDev13701766GPT-6.2 Astra
Terminal-Bench 4 (Vals AI)Not measured63.1%Only GPT-6.2 Astra measured
Vibe Code Bench (Vals AI)Not measured89.9%Only GPT-6.2 Astra measured
SciCode45.5%Not measuredOnly Gemma 4 31B measured
Terminal-Bench Hard (Artificial Analysis run)36.4%Not measuredOnly Gemma 4 31B measured
Terminal-Bench 4.0 (Artificial Analysis run)0.0%64.1%GPT-6.2 Astra
FrontierCode 2.00.0%49.0%GPT-6.2 Astra
WeirdML52.3%Not measuredOnly Gemma 4 31B measured
MirrorCodeNot measured67.8%Only GPT-6.2 Astra measured
ALE-Bench925.5Not measuredOnly Gemma 4 31B measured
LMArena Text: coding14591450Gemma 4 31B
LMArena Text: software and IT14631443Gemma 4 31B
LMArena WebDev: React13621770GPT-6.2 Astra
LMArena WebDev: HTML13801854GPT-6.2 Astra
SciCode8.9%Not measuredOnly Gemma 4 31B measured
GameBench 234.997.2GPT-6.2 Astra
Neither model has a result on the other 35 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
AA Intelligence Index4.462.1GPT-6.2 Astra
Epoch Capabilities Index144.1172.8GPT-6.2 Astra
Agentic workLike-for-likePublic boards, shown as published, kept out of the index03
Agentic work benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
Design Arena: agents building web appsNot measured1334Only GPT-6.2 Astra measured
Design Arena: agents building full-stack appsNot measured1351Only GPT-6.2 Astra measured
Design Arena: agents building mobile appsNot measured1305Only GPT-6.2 Astra measured
Design Arena: agents building Android appsNot measured1257Only GPT-6.2 Astra measured
Design Arena: spreadsheetsNot measured1278Only GPT-6.2 Astra measured
LiveBench research38.1%91.5%GPT-6.2 Astra
LiveBench browser0.0%93.0%GPT-6.2 Astra
LiveBench long agentic8.5%70.8%GPT-6.2 Astra
Tax Agent Bench (Vals AI)Not measured21.8%Only GPT-6.2 Astra measured
Legal Research Bench (Vals AI)Not measured42.8%Only GPT-6.2 Astra measured
Harvey's Legal Agent Benchmark (Vals AI)Not measured16.3%Only GPT-6.2 Astra measured
BioMysteryBench (Vals AI)Not measured80.7%Only GPT-6.2 Astra measured
Terminal-Bench-Science (Vals AI)Not measured71.4%Only GPT-6.2 Astra measured
Vals Science IndexNot measured68.4%Only GPT-6.2 Astra measured
Robo-bench (Vals AI)Not measured32.7%Only GPT-6.2 Astra measured
τ²-Bench Telecom65.5%Not measuredOnly Gemma 4 31B measured
APEX-AgentsNot measured72.2%Only GPT-6.2 Astra measured
SpireClimbNot measured41.3%Only GPT-6.2 Astra measured
LMArena SearchNot measured1191Only GPT-6.2 Astra measured
Neither model has a result on the other 25 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index24
Reasoning benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
LiveBench reasoning68.3%91.6%GPT-6.2 Astra
MysteryMechanism (Vals AI)Not measured53.1%Only GPT-6.2 Astra measured
GPQA Diamond (Artificial Analysis run)85.7%Not measuredOnly Gemma 4 31B measured
Humanity's Last Exam (Artificial Analysis run)23.6%Not measuredOnly Gemma 4 31B measured
HLE-Diamond (Artificial Analysis run)7.2%72.2%GPT-6.2 Astra
CritPt v2, hard subset (Artificial Analysis run)4.0%71.5%GPT-6.2 Astra
GPQA Diamond (Epoch AI run)75.8%96.8%GPT-6.2 Astra
ARC-AGINot measured98.5%Only GPT-6.2 Astra measured
ARC-AGI-2Not measured95.8%Only GPT-6.2 Astra measured
SimpleBenchNot measured77.3%Only GPT-6.2 Astra measured
Chess puzzlesNot measured78.0%Only GPT-6.2 Astra measured
Mystery game puzzlesNot measured95.0%Only GPT-6.2 Astra measured
EBR-benchNot measured92.4%Only GPT-6.2 Astra measured
LMArena Text: hard prompts14471425Gemma 4 31B
LMArena Text: expert1465Not measuredOnly Gemma 4 31B measured
LMArena Text: hard prompts in English14561426Gemma 4 31B
LMArena Vision: diagrams1287Not measuredOnly Gemma 4 31B measured
LMArena Vision: homework1310Not measuredOnly Gemma 4 31B measured
Kagi LLM Benchmark63.5%Not measuredOnly Gemma 4 31B measured
Humanity's Last Exam (Scale run)Not measured66.9%Only GPT-6.2 Astra measured
GoBenchNot measured2493Only GPT-6.2 Astra measured
Neither model has a result on the other 5 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index01
Math benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
OTIS Mock AIME73.3%100.0%GPT-6.2 Astra
FrontierMath, tiers 1 to 3Not measured92.6%Only GPT-6.2 Astra measured
FrontierMath, tier 4Not measured100.0%Only GPT-6.2 Astra measured
LMArena Text: math1461Not measuredOnly Gemma 4 31B measured
LMArena Text: mathematical work1464Not measuredOnly Gemma 4 31B measured
MathArenaNot measured92.9%Only GPT-6.2 Astra measured
Neither model has a result on the other 8 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index32
Knowledge benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
CaseLaw (Vals AI)52.6%Not measuredOnly Gemma 4 31B measured
MortgageTax (Vals AI)61.4%Not measuredOnly Gemma 4 31B measured
AA-Omniscience-48.550.0GPT-6.2 Astra
SimpleQA Verified10.4%78.5%GPT-6.2 Astra
LMArena Text: science14551415Gemma 4 31B
LMArena Text: medicine and healthcare1450Not measuredOnly Gemma 4 31B measured
LMArena Text: legal and government14381415Gemma 4 31B
LMArena Text: business and finance14311413Gemma 4 31B
FACTS Benchmark Suite50.1%Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 12 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index01
Web design benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
Design Arena: websitesNot measured1333Only GPT-6.2 Astra measured
Design Arena: UI componentsNot measured1341Only GPT-6.2 Astra measured
Design Arena: game devNot measured1501Only GPT-6.2 Astra measured
Design Arena: 3DNot measured1430Only GPT-6.2 Astra measured
Design Arena: all code categoriesNot measured1389Only GPT-6.2 Astra measured
LiveBench design56.5%82.8%GPT-6.2 Astra
Neither model has a result on the other 3 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index74
General benchmarks for Gemma 4 31B and GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
Vals IndexNot measured67.7%Only GPT-6.2 Astra measured
LiveBench data analysis53.3%84.1%GPT-6.2 Astra
SAGE (Vals AI)82.8%Not measuredOnly Gemma 4 31B measured
IFBench75.6%Not measuredOnly Gemma 4 31B measured
Long context reasoning (AA-LCR)69.7%Not measuredOnly Gemma 4 31B measured
GDPval-AA627Not measuredOnly Gemma 4 31B measured
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
Furniture assemblyNot measured89.2%Only GPT-6.2 Astra measured
LMArena Text14421414Gemma 4 31B
LMArena Text: instruction following14321423Gemma 4 31B
LMArena Text: creative writing14151417GPT-6.2 Astra
LMArena Vision1279Not measuredOnly Gemma 4 31B measured
LMArena Text: English14511417Gemma 4 31B
LMArena Text: non-English14301404Gemma 4 31B
LMArena Text: Chinese1476Not measuredOnly Gemma 4 31B measured
LMArena Text: French1435Not measuredOnly Gemma 4 31B measured
LMArena Text: Spanish1443Not measuredOnly Gemma 4 31B measured
LMArena Text: Russian14601437Gemma 4 31B
LMArena Text: Polish1431Not measuredOnly Gemma 4 31B measured
LMArena Text: multi-turn14521425Gemma 4 31B
LMArena Text: longer queries14451431Gemma 4 31B
LMArena Text: writing and language14311440GPT-6.2 Astra
LMArena Text: entertainment and media14041413GPT-6.2 Astra
LMArena Vision: English1274Not measuredOnly Gemma 4 31B measured
LMArena Vision: Chinese1327Not measuredOnly Gemma 4 31B measured
LMArena Vision: reading text1285Not measuredOnly Gemma 4 31B measured
LMArena Vision: humor1280Not measuredOnly Gemma 4 31B measured
LMArena Vision: captioning1210Not measuredOnly Gemma 4 31B measured
LMArena Vision: recognizing things1228Not measuredOnly Gemma 4 31B measured
LMArena Vision: creative writing1287Not measuredOnly Gemma 4 31B measured
LMArena Document1434Not measuredOnly Gemma 4 31B measured
MultiChallengeNot measured73.2%Only GPT-6.2 Astra measured
EQ-Bench 41110Not measuredOnly Gemma 4 31B measured
Neither model has a result on the other 14 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 GPT-6.2 Astra
BenchmarkGemma 4 31BGPT-6.2 AstraLeader
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$0.0003Fits in one request
GPT-6.2 Astra$0.028Fits in one request

Gemma 4 31B costs less

Repository review

50K input + 3K output tokens

Gemma 4 31B$0.0082Fits in one request
GPT-6.2 Astra$0.520Fits in one request

Gemma 4 31B costs less

Agent loop

200K input + 10K output tokens

Gemma 4 31B$0.032Fits in one request
GPT-6.2 Astra$2.00Fits in one request

Gemma 4 31B 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 GPT-6.2 Astra
SpecificationGoogle Gemma 4 31BOpenAI GPT-6.2 Astra
MakerGoogleOpenAI
WeightsOpen weightsNo public weights listed
ReleasedApril 2, 2026November 17, 2026
Context window262K1.05M
Input $/Mtok$0.14$8.00
Output $/Mtok$0.40$40.00
Coding index21.987.0
Independent measurements behind it24
Cost per solved taskNot computableNot computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Gemma 4 31B better than GPT-6.2 Astra for coding?

Pick GPT-6.2 Astra for coding: it scores 87.0 on the Vibe Coding Index against 21.9 for Gemma 4 31B. Pick Gemma 4 31B for output tokens 100.0x cheaper ($0.40 vs $40.00 per million) and work measured by LMArena Text: coding or LMArena Text: software and IT, where it leads.

Which is cheaper, Gemma 4 31B or GPT-6.2 Astra?

Gemma 4 31B costs $0.14 per million input tokens and $0.40 per million output tokens; GPT-6.2 Astra costs $8.00 and $40.00. Cost per solved task is known for neither of them.

Which has the bigger context window, Gemma 4 31B or GPT-6.2 Astra?

GPT-6.2 Astra, with 1.05M tokens against 262K.

How many benchmarks is the Gemma 4 31B vs GPT-6.2 Astra comparison based on?

38 benchmarks have a result for both models as of November 22, 2026. Gemma 4 31B's coding index rests on 2 independent benchmarks and GPT-6.2 Astra's on 4. Where only one model has a result, the page says so instead of naming a winner.

Compare next

Embed this comparison on your site

Free under CC BY 4.0, while the source link stays visible. Preview the frame · Terms and other embeds

<iframe src="https://llmbenchmark.ai/embed/compare/gemma-4-31b-it-vs-gpt-6-2-astra" title="Gemma 4 31B vs GPT-6.2 Astra, LLMBenchmark.ai" width="100%" height="360" style="border:0;max-width:100%" loading="lazy"></iframe>
<p style="font:13px/1.4 sans-serif;margin:6px 0 0">Source: <a href="https://llmbenchmark.ai/compare/gemma-4-31b-it-vs-gpt-6-2-astra">LLMBenchmark.ai, Gemma 4 31B vs GPT-6.2 Astra</a></p>