Gemma 3 4B vs Grok 4.7
Grok 4.7 leads Gemma 3 4B on the Artificial Analysis Intelligence Index, 46.4 to 4.8, and Gemma 3 4B costs 80.0x less per output token ($0.075 vs $6.00 per million). Updated October 22, 2026; 30 benchmarks have a result for both.
Gemma 3 4B or Grok 4.7: which to pick
Only Grok 4.7 has a coding score, 66.7 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding.
Pick Gemma 3 4B for output tokens 80.0x cheaper ($0.075 vs $6.00 per million).
- Shared results
- 30
- Gemma 3 4B only
- 12
- Grok 4.7 only
- 69
- Like-for-like categories
- 6 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.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only Grok 4.7 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Grok 4.7
Grok 4.7 leads, 46.4 against 4.8.
Same board, both measured
Web design
Design Arena: websites
No call
Only Grok 4.7 has a Design Arena: websites result, 1255.
Directional only
General
Vals Index
No call
Only Grok 4.7 has a Vals Index result, 58.6%.
Directional only
Long documents
Listed context window
Grok 4.7
Grok 4.7 takes 500K tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemma 3 4B
Gemma 3 4B $0.0001, Grok 4.7 $0.0050 per turn.
List prices
At a glance
- Coding index
- Gemma 3 4B not benchmarked, Grok 4.7 66.7 from 5 benchmarks.
- Widest category
- Math, 49.8 points apart on average over 3 shared benchmarks.
- Widest single gap
- OTIS Mock AIME: Gemma 3 4B 7.5% against Grok 4.7 98.1%, 90.6 points apart on a 0 to 100 scale.
- Output price
- Gemma 3 4B $0.075 against Grok 4.7 $6.00 per million output tokens, 80.0x apart.
- Context window
- Gemma 3 4B takes 131K tokens, Grok 4.7 500K.
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.
| Category | Gemma 3 4B | Bars | Grok 4.7 | Ties |
|---|---|---|---|---|
| Coding | 0 | 2 | ||
| Intelligence | 0 | 2 | ||
| Reasoning | 0 | 5 | ||
| Math | 0 | 3 | ||
| Knowledge | 0 | 4 | ||
| General | 0 | 14 | ||
| All shared boards | 0 | 30 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index02
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 84.5% | Only Grok 4.7 measured |
| LMArena WebDev | Not measured | 1637 | Only Grok 4.7 measured |
| AA Coding Index | 2.7 | Not measured | Only Gemma 3 4B measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 28.8% | Only Grok 4.7 measured |
| Vibe Code Bench (Vals AI) | Not measured | 86.2% | Only Grok 4.7 measured |
| Code migration (Vals AI) | Not measured | 44.8% | Only Grok 4.7 measured |
| ProgramBench (Vals AI) | Not measured | 0.5% | Only Grok 4.7 measured |
| IOI (Vals AI) | Not measured | 57.7% | Only Grok 4.7 measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 73.4% | Only Grok 4.7 measured |
| SWE-bench Multimodal (Vals AI) | Not measured | 42.9% | Only Grok 4.7 measured |
| CyberBench (Vals AI) | Not measured | 69.5% | Only Grok 4.7 measured |
| LiveCodeBench (Artificial Analysis run) | 11.2% | Not measured | Only Gemma 3 4B measured |
| SciCode | Not measured | 57.8% | Only Grok 4.7 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 0.8% | Not measured | Only Gemma 3 4B measured |
| FrontierCode | Not measured | 47.6% | Only Grok 4.7 measured |
| FrontierSWE | Not measured | 29.5% | Only Grok 4.7 measured |
| LMArena Text: coding | 1230 | 1432 | Grok 4.7 |
| LMArena Text: software and IT | 1269 | 1421 | Grok 4.7 |
| LMArena WebDev: React | Not measured | 1663 | Only Grok 4.7 measured |
| LMArena WebDev: HTML | Not measured | 1627 | Only Grok 4.7 measured |
| LMArena WebDev: from an image | Not measured | 1586 | Only Grok 4.7 measured |
| SciCode | 0.0% | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 17 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| AA Intelligence Index | 4.8 | 46.4 | Grok 4.7 |
| Epoch Capabilities Index | 115.9 | 154.4 | Grok 4.7 |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index17 boards, none shared
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| LiveBench research | Not measured | 86.4% | Only Grok 4.7 measured |
| LiveBench browser | Not measured | 60.0% | Only Grok 4.7 measured |
| LiveBench long agentic | Not measured | 50.1% | Only Grok 4.7 measured |
| SREBench (Vals AI) | Not measured | 3.8% | Only Grok 4.7 measured |
| Finance Agent (Vals AI) | Not measured | 52.3% | Only Grok 4.7 measured |
| Tax Agent Bench (Vals AI) | Not measured | 34.2% | Only Grok 4.7 measured |
| Legal Research Bench (Vals AI) | Not measured | 47.1% | Only Grok 4.7 measured |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 18.3% | Only Grok 4.7 measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 67.0% | Only Grok 4.7 measured |
| BioMysteryBench (Vals AI) | Not measured | 69.3% | Only Grok 4.7 measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 10.0% | Only Grok 4.7 measured |
| Vals Science Index | Not measured | 34.8% | Only Grok 4.7 measured |
| Robo-bench (Vals AI) | Not measured | 0.2% | Only Grok 4.7 measured |
| τ²-Bench Telecom | 5.0% | Not measured | Only Gemma 3 4B measured |
| APEX-Agents | Not measured | 54.6% | Only Grok 4.7 measured |
| Vending-Bench 2 | Not measured | $10,537 | Only Grok 4.7 measured |
| Berkeley Function Calling Leaderboard (BFCL) | 19.6% | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 24 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| LiveBench reasoning | Not measured | 82.7% | Only Grok 4.7 measured |
| MysteryMechanism (Vals AI) | Not measured | 25.2% | Only Grok 4.7 measured |
| GPQA Diamond (Artificial Analysis run) | 29.1% | Not measured | Only Gemma 3 4B measured |
| Humanity's Last Exam (Artificial Analysis run) | 5.3% | 43.1% | Grok 4.7 |
| GPQA Diamond (Epoch AI run) | 23.2% | 92.7% | Grok 4.7 |
| ARC-AGI | Not measured | 90.2% | Only Grok 4.7 measured |
| ARC-AGI-2 | Not measured | 61.4% | Only Grok 4.7 measured |
| Chess puzzles | Not measured | 41.0% | Only Grok 4.7 measured |
| Mystery game puzzles | Not measured | 29.0% | Only Grok 4.7 measured |
| LMArena Text: hard prompts | 1253 | 1415 | Grok 4.7 |
| LMArena Text: expert | 1223 | 1439 | Grok 4.7 |
| LMArena Text: hard prompts in English | 1253 | 1427 | Grok 4.7 |
| LMArena Vision: diagrams | Not measured | 1254 | Only Grok 4.7 measured |
| LMArena Vision: homework | Not measured | 1309 | Only Grok 4.7 measured |
| Kagi LLM Benchmark | 25.2% | Not measured | Only Gemma 3 4B measured |
| Neither model has a result on the other 7 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| AA Math Index | 12.7 | Not measured | Only Gemma 3 4B measured |
| AIME 2025 (Artificial Analysis run) | 12.7% | Not measured | Only Gemma 3 4B measured |
| OTIS Mock AIME | 7.5% | 98.1% | Grok 4.7 |
| FrontierMath, tiers 1 to 3 | Not measured | 53.0% | Only Grok 4.7 measured |
| FrontierMath, tier 4 | Not measured | 17.1% | Only Grok 4.7 measured |
| ProofBench | Not measured | 26.0% | Only Grok 4.7 measured |
| LMArena Text: math | 1239 | 1414 | Grok 4.7 |
| LMArena Text: mathematical work | 1256 | 1435 | Grok 4.7 |
| Neither model has a result on the other 5 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| LegalBench (Vals AI) | Not measured | 84.4% | Only Grok 4.7 measured |
| MedCode (Vals AI) | Not measured | 49.5% | Only Grok 4.7 measured |
| MMLU Pro (Artificial Analysis run) | 41.7% | Not measured | Only Gemma 3 4B measured |
| SimpleQA Verified | Not measured | 56.0% | Only Grok 4.7 measured |
| LMArena Text: science | 1306 | 1419 | Grok 4.7 |
| LMArena Text: medicine and healthcare | 1342 | 1395 | Grok 4.7 |
| LMArena Text: legal and government | 1327 | 1423 | Grok 4.7 |
| LMArena Text: business and finance | 1311 | 1389 | Grok 4.7 |
| 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
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1255 | Only Grok 4.7 measured |
| Design Arena: UI components | Not measured | 1275 | Only Grok 4.7 measured |
| Design Arena: game dev | Not measured | 1314 | Only Grok 4.7 measured |
| Design Arena: data viz | Not measured | 1238 | Only Grok 4.7 measured |
| Design Arena: 3D | Not measured | 1274 | Only Grok 4.7 measured |
| Design Arena: all code categories | Not measured | 1273 | Only Grok 4.7 measured |
| LiveBench design | Not measured | 78.0% | Only Grok 4.7 measured |
| Neither model has a result on the other 2 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index014
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| Vals Index | Not measured | 58.6% | Only Grok 4.7 measured |
| LiveBench data analysis | Not measured | 76.9% | Only Grok 4.7 measured |
| Public Benefits Bench (Vals AI) | Not measured | 65.6% | Only Grok 4.7 measured |
| SAGE (Vals AI) | Not measured | 40.8% | Only Grok 4.7 measured |
| MedScribe (Vals AI) | Not measured | 89.4% | Only Grok 4.7 measured |
| IFBench | 28.3% | Not measured | Only Gemma 3 4B measured |
| Long context reasoning (AA-LCR) | 6.7% | 80.0% | Grok 4.7 |
| DTBench | 50.9% | 96.0% | Grok 4.7 |
| LMCA | 2.8% | 49.4% | Grok 4.7 |
| Furniture assembly | Not measured | 20.8% | Only Grok 4.7 measured |
| LMArena Text | 1291 | 1399 | Grok 4.7 |
| LMArena Text: instruction following | 1239 | 1406 | Grok 4.7 |
| LMArena Text: creative writing | 1270 | 1381 | Grok 4.7 |
| LMArena Vision | Not measured | 1231 | Only Grok 4.7 measured |
| LMArena Text: English | 1300 | 1411 | Grok 4.7 |
| LMArena Text: non-English | 1273 | 1386 | Grok 4.7 |
| LMArena Text: Chinese | Not measured | 1449 | Only Grok 4.7 measured |
| LMArena Text: German | 1281 | 1366 | Grok 4.7 |
| LMArena Text: Japanese | Not measured | 1359 | Only Grok 4.7 measured |
| LMArena Text: Korean | Not measured | 1383 | Only Grok 4.7 measured |
| LMArena Text: French | Not measured | 1431 | Only Grok 4.7 measured |
| LMArena Text: Spanish | Not measured | 1406 | Only Grok 4.7 measured |
| LMArena Text: Russian | 1294 | 1399 | Grok 4.7 |
| LMArena Text: Polish | Not measured | 1420 | Only Grok 4.7 measured |
| LMArena Text: multi-turn | 1255 | 1394 | Grok 4.7 |
| LMArena Text: longer queries | 1273 | 1416 | Grok 4.7 |
| LMArena Text: writing and language | 1288 | 1394 | Grok 4.7 |
| LMArena Text: entertainment and media | 1233 | 1376 | Grok 4.7 |
| LMArena Vision: English | Not measured | 1229 | Only Grok 4.7 measured |
| LMArena Vision: Chinese | Not measured | 1250 | Only Grok 4.7 measured |
| LMArena Vision: reading text | Not measured | 1241 | Only Grok 4.7 measured |
| LMArena Vision: creative writing | Not measured | 1217 | Only Grok 4.7 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
| Benchmark | Gemma 3 4B | Grok 4.7 | Leader |
|---|---|---|---|
| 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 costs less
Repository review
50K input + 3K output tokens
Gemma 3 4B costs less
Agent loop
200K input + 10K output tokens
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.
| Specification | Google Gemma 3 4B | SpaceXAI Grok 4.7 |
|---|---|---|
| Maker | xAI | |
| Weights | Open weights | No public weights listed |
| Released | March 13, 2025 | September 21, 2026 |
| Context window | 131K | 500K |
| Input $/Mtok | $0.045 | $2.00 |
| Output $/Mtok | $0.075 | $6.00 |
| Coding index | Not benchmarked | 66.7 |
| Independent measurements behind it | 0 | 5 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemma 3 4B better than Grok 4.7 for coding?
Only Grok 4.7 has a coding score, 66.7 on the Vibe Coding Index; Gemma 3 4B has none, so the two cannot be ranked on coding. Pick Gemma 3 4B for output tokens 80.0x cheaper ($0.075 vs $6.00 per million).
Which is cheaper, Gemma 3 4B or Grok 4.7?
Gemma 3 4B costs $0.045 per million input tokens and $0.075 per million output tokens; Grok 4.7 costs $2.00 and $6.00. Cost per solved task is known for neither of them.
Which has the bigger context window, Gemma 3 4B or Grok 4.7?
Grok 4.7, with 500K tokens against 131K.
How many benchmarks is the Gemma 3 4B vs Grok 4.7 comparison based on?
30 benchmarks have a result for both models as of October 22, 2026. Gemma 3 4B's coding index rests on 0 independent benchmarks and Grok 4.7's on 5. Where only one model has a result, the page says so instead of naming a winner.