Model comparison
DeepSeek-R1 vs Gemma 4 31B IT
Gemma 4 31B IT is the stronger model overall, scoring 43.2 to 42.0 on the Noometry Index.
Last verified . 24 shared benchmarks.
DeepSeek
GoogleSummary
- They share 24 benchmarks with published results for both. DeepSeek-R1 scores higher in 5 categories and Gemma 4 31B IT in 3 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Gemma 4 31B IT leads 26.4 to 18.7.
- The biggest single-benchmark swing is WeirdML: 41.6% for DeepSeek-R1 and 52.3% for Gemma 4 31B IT.
- Gemma 4 31B IT is cheaper at $0.09 / $0.34 per million input/output tokens, against $0.50 / $2.15 for DeepSeek-R1.
- Gemma 4 31B IT accepts more context: 262K tokens versus 164K.
- Gemma 4 31B IT has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-R1 | Gemma 4 31B IT | |
|---|---|---|
| Provider | DeepSeek | |
| Noometry Index | 42.0 | 43.2 |
| Released | 2025-01-20 | 2026-04-02 |
| Weights | Proprietary | Open |
| Context window | 164K | 262K |
| Max output | 64K | 33K |
| Input $ / M tokens | $0.50 | $0.09 |
| Output $ / M tokens | $2.15 | $0.34 |
| Results tracked | 52 | 35 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-R1 leads
DeepSeek-R1: 45.9 (#63), Gemma 4 31B IT: 41.9 (#98)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| SciCode | 35.7% | 43.4% |
| WeirdML | 41.6% | 52.3% |
| LMArena Coding | 1427 | 1459 |
| ALE-Bench | 804.12 | 925.5 |
| Aider Polyglot | 71.4% | — |
| LMArena WebDev | — | 1364 |
| LiveBench Coding | 66.7% | — |
| AlgoTune | 1.7 | — |
Agentic & Tool Use Not comparable
DeepSeek-R1: 30.4 (#68), Gemma 4 31B IT: —
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| DeepResearch Bench | 35.1% | — |
| BALROG | 34.9% | — |
| METR Time Horizons | 53.8% | — |
Reasoning Gemma 4 31B IT leads
DeepSeek-R1: 18.7 (#253), Gemma 4 31B IT: 26.4 (#119)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| Kagi LLM Benchmark | 69.4% | 63.5% |
| CritPt | 1.1% | 1.4% |
| LMArena Hard Prompts | 1415 | 1448 |
| Epoch Capabilities Index | 141.3 | 142.74 |
| ARC-AGI-2 | 1.3% | — |
| SimpleBench | 40.8% | — |
| NYT Connections (extended) | — | 70.6% |
| ARC-AGI-1 | 21.2% | — |
| Chess Puzzles | — | 5% |
| Thematic Generalization | — | 53% |
| LiveBench Reasoning | 83.2% | — |
| DTBench | — | 82.7% |
| LiveBench Data Analysis | 69.8% | — |
| LMCA | — | 39.3% |
| Surface Evolver Bench | — | 30.6% |
| ForecastBench | 60 | — |
| LiveBench | 71.6% | — |
Math Too close to call
DeepSeek-R1: 42.0 (#70), Gemma 4 31B IT: 41.3 (#72)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.4% | 73.3% |
| LMArena Math | 1401 | 1464 |
| Omni-MATH | 42.4% | — |
| LiveBench Math | 80.7% | — |
| MATH Level 5 | 96.6% | — |
Knowledge DeepSeek-R1 leads
DeepSeek-R1: 44.3 (#81), Gemma 4 31B IT: 37.9 (#141)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| GPQA Diamond | 76.3% | 75.8% |
| Vectara Hallucination Rate | 11.3% | 7.4% |
| LMArena Expert | 1394 | 1465 |
| SimpleQA Verified | — | 10.4% |
| MMLU-Pro | 79.3% | — |
| Confabulations | 12.7% | — |
| GPQA (HELM) | 66.6% | — |
Multimodal Not comparable
DeepSeek-R1: —, Gemma 4 31B IT: 41.0 (#32)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| LMArena Vision | — | 1278 |
| LMArena Document | — | 1425 |
Multilingual Gemma 4 31B IT leads
DeepSeek-R1: 52.6 (#80), Gemma 4 31B IT: 54.0 (#52)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| LMArena Non-English | 1412 | 1431 |
| LMArena Chinese | 1443 | 1478 |
| LMArena French | 1415 | 1436 |
| LMArena Russian | 1423 | 1462 |
| LMArena Spanish | 1410 | 1443 |
| LMArena German | 1405 | — |
| LMArena Japanese | 1394 | — |
| LMArena Korean | 1359 | — |
Instruction Following Gemma 4 31B IT leads
DeepSeek-R1: 72.2 (#138), Gemma 4 31B IT: 75.6 (#58)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| LMArena Instruction Following | 1382 | 1433 |
| LiveBench Instruction Following | 80.5% | — |
| IFEval | 78.4% | — |
Long Context DeepSeek-R1 leads
DeepSeek-R1: 45.4 (#36), Gemma 4 31B IT: 44.2 (#67)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| LMArena Longer Query | 1391 | 1446 |
| Fiction.LiveBench | 75% | — |
Writing & Preference Too close to call
DeepSeek-R1: 61.5 (#84), Gemma 4 31B IT: 60.7 (#92)
| Benchmark | DeepSeek-R1 | Gemma 4 31B IT |
|---|---|---|
| LMArena Text | 1428 | 1443 |
| LMArena Creative Writing | 1406 | 1416 |
| EQ-Bench Creative Writing | 1500 | 1368 |
| LMArena Multi-Turn | 1404 | 1452 |
| Short-Story Creative Writing | 83% | — |
| WildBench | 82.8% | — |
| EQ-Bench 4 | — | 1120 |
| LiveBench Language | 48.5% | — |
Frequently asked questions
Is DeepSeek-R1 better than Gemma 4 31B IT?
Gemma 4 31B IT is the stronger model overall, scoring 43.2 to 42.0 on the Noometry Index.
Which is cheaper, DeepSeek-R1 or Gemma 4 31B IT?
Gemma 4 31B IT is cheaper. It lists at $0.09 per million input tokens and $0.34 per million output tokens; DeepSeek-R1 lists at $0.50 and $2.15.
Is DeepSeek-R1 or Gemma 4 31B IT better for coding?
DeepSeek-R1 scores higher on coding benchmarks: 45.9 versus 41.9 in the Noometry coding category.
Which has the bigger context window?
Gemma 4 31B IT does, with 262K tokens against 164K.
How many benchmarks do DeepSeek-R1 and Gemma 4 31B IT share?
24 benchmarks have published results for both models. DeepSeek-R1 has 52 scored results on Noometry and Gemma 4 31B IT has 35.