Model comparison
Gemma 3 4B vs GPT-5.6 Terra
GPT-5.6 Terra is the stronger model overall, scoring 59.1 to 27.9 on the Noometry Index. Gemma 3 4B costs 90× less per token, which makes it the better buy when GPT-5.6 Terra's lead doesn't matter for your workload.
Last verified . 20 shared benchmarks.
GoogleSummary
- They share 20 benchmarks with published results for both. Gemma 3 4B scores higher in 0 categories and GPT-5.6 Terra in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GPT-5.6 Terra leads 81.5 to 15.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 7.5% for Gemma 3 4B and 99.7% for GPT-5.6 Terra.
- Gemma 3 4B is cheaper at $0.04 / $0.08 per million input/output tokens, against $2 / $12 for GPT-5.6 Terra.
- GPT-5.6 Terra accepts more context: 1.05M tokens versus 131K.
- Gemma 3 4B has downloadable open weights; the other is API-only.
Side by side
| Gemma 3 4B | GPT-5.6 Terra | |
|---|---|---|
| Provider | OpenAI | |
| Noometry Index | 27.9 | 59.1 |
| Released | 2025-03-12 | 2026-07-09 |
| Weights | Open | Proprietary |
| Context window | 131K | 1.05M |
| Max output | 4K | 128K |
| Input $ / M tokens | $0.04 | $2 |
| Output $ / M tokens | $0.08 | $12 |
| Results tracked | 22 | 52 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.6 Terra leads
Gemma 3 4B: 35.3 (#214), GPT-5.6 Terra: 57.1 (#16)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Coding | 1230 | 1484 |
| DeepSWE | — | 69.6% |
| FrontierCode | — | 41.3% |
| CursorBench | — | 41.3% |
| LMArena WebDev | — | 1520 |
| SciCode | — | 55% |
| WeirdML | — | 78.3% |
| ALE-Bench | — | 1,951 |
Agentic & Tool Use GPT-5.6 Terra leads
Gemma 3 4B: 20.7 (#134), GPT-5.6 Terra: 40.2 (#23)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| APEX-Agents | — | 58.2% |
| Berkeley Function Calling Leaderboard | 19.6% | — |
| BALROG | — | 53.2% |
| GDP.pdf | — | 24.7% |
| Vending-Bench 2 | — | 7,343 |
Reasoning GPT-5.6 Terra leads
Gemma 3 4B: 12.7 (#330), GPT-5.6 Terra: 60.6 (#19)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| Kagi LLM Benchmark | 25.2% | 51.3% |
| Chess Puzzles | 0% | 54% |
| LMArena Hard Prompts | 1253 | 1467 |
| DTBench | 50.9% | 93.3% |
| LMCA | 2.8% | 55% |
| Epoch Capabilities Index | 116.02 | 159.31 |
| ARC-AGI-2 | — | 83.9% |
| SimpleBench | — | 48.9% |
| NYT Connections (extended) | — | 78.4% |
| ARC-AGI-1 | — | 96.5% |
| CritPt | — | 30% |
| Mystery Game Puzzles | — | 35% |
| Surface Evolver Bench | — | 83.8% |
Math GPT-5.6 Terra leads
Gemma 3 4B: 15.8 (#286), GPT-5.6 Terra: 81.5 (#10)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 7.5% | 99.7% |
| LMArena Math | 1239 | 1463 |
| FrontierMath (Tiers 1-3) | — | 86% |
| FrontierMath Tier 4 | — | 70.7% |
| ProofBench | — | 74% |
Knowledge GPT-5.6 Terra leads
Gemma 3 4B: 11.8 (#294), GPT-5.6 Terra: 61.3 (#27)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| GPQA Diamond | 23.2% | 93.3% |
| LMArena Expert | 1223 | 1492 |
| SimpleQA Verified | — | 43.2% |
| Vectara Hallucination Rate | 6.4% | — |
Multimodal Not comparable
Gemma 3 4B: —, GPT-5.6 Terra: 47.5 (#10)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Vision | — | 1271 |
| Blueprint-Bench 2 | — | 30.8% |
| Furniture Assembly | — | 54.2% |
| LMArena Document | — | 1472 |
Multilingual GPT-5.6 Terra leads
Gemma 3 4B: 42.7 (#189), GPT-5.6 Terra: 54.5 (#42)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Non-English | 1273 | 1439 |
| LMArena German | 1281 | 1462 |
| LMArena Russian | 1294 | 1450 |
| LMArena Chinese | — | 1514 |
| LMArena French | — | 1476 |
| LMArena Japanese | — | 1446 |
| LMArena Korean | — | 1420 |
| LMArena Spanish | — | 1455 |
Instruction Following GPT-5.6 Terra leads
Gemma 3 4B: 65.4 (#219), GPT-5.6 Terra: 76.6 (#38)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Instruction Following | 1239 | 1453 |
Long Context GPT-5.6 Terra leads
Gemma 3 4B: 38.6 (#189), GPT-5.6 Terra: 44.4 (#64)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Longer Query | 1274 | 1452 |
Writing & Preference GPT-5.6 Terra leads
Gemma 3 4B: 42.2 (#234), GPT-5.6 Terra: 70.4 (#21)
| Benchmark | Gemma 3 4B | GPT-5.6 Terra |
|---|---|---|
| LMArena Text | 1291 | 1446 |
| LMArena Creative Writing | 1270 | 1411 |
| EQ-Bench Creative Writing | 1068 | 1855 |
| LMArena Multi-Turn | 1255 | 1451 |
| EQ-Bench 4 | — | 1234 |
Frequently asked questions
Is Gemma 3 4B better than GPT-5.6 Terra?
GPT-5.6 Terra is the stronger model overall, scoring 59.1 to 27.9 on the Noometry Index. Gemma 3 4B costs 90× less per token, which makes it the better buy when GPT-5.6 Terra's lead doesn't matter for your workload.
Which is cheaper, Gemma 3 4B or GPT-5.6 Terra?
Gemma 3 4B is cheaper. It lists at $0.04 per million input tokens and $0.08 per million output tokens; GPT-5.6 Terra lists at $2 and $12.
Is Gemma 3 4B or GPT-5.6 Terra better for coding?
GPT-5.6 Terra scores higher on coding benchmarks: 57.1 versus 35.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5.6 Terra does, with 1.05M tokens against 131K.
How many benchmarks do Gemma 3 4B and GPT-5.6 Terra share?
20 benchmarks have published results for both models. Gemma 3 4B has 22 scored results on Noometry and GPT-5.6 Terra has 52.