Model comparison
Gemma 3 4B vs GPT-4o mini
Gemma 3 4B is the stronger model overall, scoring 28.1 to 25.5 on the Noometry Index.
Last verified . 20 shared benchmarks.
GoogleSummary
- They share 20 benchmarks with published results for both. Gemma 3 4B scores higher in 6 categories and GPT-4o mini in 3 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in coding, where Gemma 3 4B leads 35.9 to 22.0.
- The biggest single-benchmark swing is GPQA Diamond: 23.2% for Gemma 3 4B and 37.7% for GPT-4o mini.
- Gemma 3 4B is cheaper at $0.04 / $0.08 per million input/output tokens, against $0.15 / $0.60 for GPT-4o mini.
- Gemma 3 4B accepts more context: 131K tokens versus 128K.
- Gemma 3 4B has downloadable open weights; the other is API-only.
Side by side
| Gemma 3 4B | GPT-4o mini | |
|---|---|---|
| Provider | OpenAI | |
| Noometry Index | 28.1 | 25.5 |
| Released | 2025-03-12 | 2024-07-18 |
| Weights | Open | Proprietary |
| Context window | 131K | 128K |
| Max output | 4K | 16K |
| Input $ / M tokens | $0.04 | $0.15 |
| Output $ / M tokens | $0.08 | $0.60 |
| Results tracked | 22 | 60 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Gemma 3 4B leads
Gemma 3 4B: 35.9 (#215), GPT-4o mini: 22.0 (#335)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Coding | 1230 | 1290 |
| Aider Polyglot | — | 3.6% |
| WeirdML | — | 11.8% |
| BigCodeBench Instruct | — | 46.1% |
| LiveBench Coding | — | 43.1% |
| BigCodeBench Complete | — | 57.4% |
| HumanEval+ | — | 83.5% |
| MBPP+ | — | 72.2% |
Agentic & Tool Use GPT-4o mini leads
Gemma 3 4B: 20.9 (#142), GPT-4o mini: 27.5 (#101)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| Berkeley Function Calling Leaderboard | 19.6% | — |
| BALROG | — | 17.4% |
Reasoning Gemma 3 4B leads
Gemma 3 4B: 13.2 (#335), GPT-4o mini: 8.7 (#347)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| Kagi LLM Benchmark | 25.2% | 28.8% |
| Chess Puzzles | 0% | 0% |
| LMArena Hard Prompts | 1253 | 1267 |
| DTBench | 50.9% | 54.4% |
| LMCA | 2.8% | 10.4% |
| Epoch Capabilities Index | 116.02 | 126.56 |
| ARC-AGI-2 | — | 0% |
| SimpleBench | — | 10.7% |
| LiveBench Reasoning | — | 32.8% |
| Mystery Game Puzzles | — | 12% |
| LiveBench Data Analysis | — | 50% |
| LiveBench | — | 41.3% |
| PIQA | — | 88.7% |
Math Gemma 3 4B leads
Gemma 3 4B: 16.8 (#292), GPT-4o mini: 10.4 (#314)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 7.5% | 6.9% |
| LMArena Math | 1239 | 1267 |
| FrontierMath (Tiers 1-3) | — | 0.7% |
| Omni-MATH | — | 28% |
| LiveBench Math | — | 36.3% |
| MATH Level 5 | — | 52.6% |
| GSM8K | — | 91.3% |
Knowledge GPT-4o mini leads
Gemma 3 4B: 11.8 (#299), GPT-4o mini: 17.7 (#284)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| GPQA Diamond | 23.2% | 37.7% |
| LMArena Expert | 1223 | 1235 |
| SimpleQA Verified | — | 8.3% |
| MMLU-Pro | — | 60.3% |
| Confabulations | — | 37.2% |
| Vectara Hallucination Rate | 6.4% | — |
| GPQA (HELM) | — | 36.8% |
| BoolQ | — | 88.7% |
| MMLU | — | 81.8% |
Multimodal Not comparable
Gemma 3 4B: —, GPT-4o mini: 25.9 (#122)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Vision | — | 1066 |
| Video-MME | — | 64.8% |
| GeoBench | — | 64% |
| VPCT | — | 34% |
Multilingual Too close to call
Gemma 3 4B: 42.5 (#194), GPT-4o mini: 42.0 (#199)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Non-English | 1273 | 1266 |
| LMArena German | 1281 | 1272 |
| LMArena Russian | 1294 | 1275 |
| LMArena Chinese | — | 1265 |
| LMArena French | — | 1297 |
| LMArena Japanese | — | 1216 |
| LMArena Korean | — | 1195 |
| LMArena Spanish | — | 1276 |
Instruction Following Gemma 3 4B leads
Gemma 3 4B: 65.2 (#225), GPT-4o mini: 61.9 (#239)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Instruction Following | 1239 | 1258 |
| LiveBench Instruction Following | — | 56.8% |
| IFEval | — | 78.2% |
Long Context Too close to call
Gemma 3 4B: 38.7 (#194), GPT-4o mini: 39.1 (#186)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Longer Query | 1273 | 1289 |
Writing & Preference Gemma 3 4B leads
Gemma 3 4B: 42.0 (#239), GPT-4o mini: 39.5 (#248)
| Benchmark | Gemma 3 4B | GPT-4o mini |
|---|---|---|
| LMArena Text | 1291 | 1286 |
| LMArena Creative Writing | 1271 | 1268 |
| EQ-Bench Creative Writing | 1068 | 873 |
| LMArena Multi-Turn | 1255 | 1285 |
| Short-Story Creative Writing | — | 67.2% |
| WildBench | — | 79.1% |
| LiveBench Language | — | 28.6% |
Frequently asked questions
Is Gemma 3 4B better than GPT-4o mini?
Gemma 3 4B is the stronger model overall, scoring 28.1 to 25.5 on the Noometry Index.
Which is cheaper, Gemma 3 4B or GPT-4o mini?
Gemma 3 4B is cheaper. It lists at $0.04 per million input tokens and $0.08 per million output tokens; GPT-4o mini lists at $0.15 and $0.60.
Is Gemma 3 4B or GPT-4o mini better for coding?
Gemma 3 4B scores higher on coding benchmarks: 35.9 versus 22.0 in the Noometry coding category.
Which has the bigger context window?
Gemma 3 4B does, with 131K tokens against 128K.
How many benchmarks do Gemma 3 4B and GPT-4o mini share?
20 benchmarks have published results for both models. Gemma 3 4B has 22 scored results on Noometry and GPT-4o mini has 60.