Model comparison
Deepseek Coder v2 vs Step 3
Step 3 is the stronger model overall, scoring 40.5 to 35.9 on the Noometry Index.
Last verified . 15 shared benchmarks.
DeepSeek
StepFunSummary
- They share 15 benchmarks with published results for both. Deepseek Coder v2 scores higher in 0 categories and Step 3 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Step 3 leads 54.3 to 38.2.
Side by side
| Deepseek Coder v2 | Step 3 | |
|---|---|---|
| Provider | DeepSeek | StepFun |
| Noometry Index | 35.9 | 40.5 |
| Released | 2024-06-17 | — |
| Weights | Open | Open |
| Context window | — | — |
| Max output | — | — |
| Input $ / M tokens | — | — |
| Output $ / M tokens | — | — |
| Results tracked | 24 | 17 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Step 3 leads
Deepseek Coder v2: 38.1 (#183), Step 3: 40.1 (#147)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Coding | 1251 | 1367 |
| BigCodeBench Instruct | 48.2% | — |
| BigCodeBench Complete | 59.7% | — |
| HumanEval+ | 82.3% | — |
| MBPP+ | 75.1% | — |
Reasoning Step 3 leads
Deepseek Coder v2: 23.6 (#176), Step 3: 28.4 (#105)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Hard Prompts | 1207 | 1355 |
| Kagi LLM Benchmark | — | 62.3% |
| WinoGrande | 83.7% | — |
Math Step 3 leads
Deepseek Coder v2: 34.9 (#190), Step 3: 37.6 (#148)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Math | 1241 | 1366 |
| GSM8K | 94.5% | — |
Knowledge Step 3 leads
Deepseek Coder v2: 32.3 (#212), Step 3: 36.8 (#164)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Expert | 1181 | 1333 |
| ARC (AI2) Challenge | 64.3% | — |
Multimodal Not comparable
Deepseek Coder v2: —, Step 3: 35.5 (#86)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Vision | — | 1177 |
Multilingual Step 3 leads
Deepseek Coder v2: 36.3 (#240), Step 3: 46.3 (#159)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Non-English | 1182 | 1327 |
| LMArena Chinese | 1201 | 1397 |
| LMArena German | 1164 | 1371 |
| LMArena Korean | 1104 | 1269 |
| LMArena Russian | 1188 | 1331 |
| LMArena Spanish | 1153 | 1371 |
| LMArena French | 1185 | — |
| LMArena Japanese | 1126 | — |
Instruction Following Step 3 leads
Deepseek Coder v2: 61.7 (#242), Step 3: 70.4 (#164)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Instruction Following | 1180 | 1332 |
Long Context Step 3 leads
Deepseek Coder v2: 37.0 (#224), Step 3: 40.3 (#157)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Longer Query | 1219 | 1326 |
Writing & Preference Step 3 leads
Deepseek Coder v2: 38.2 (#253), Step 3: 54.3 (#151)
| Benchmark | Deepseek Coder v2 | Step 3 |
|---|---|---|
| LMArena Text | 1191 | 1350 |
| LMArena Creative Writing | 1120 | 1321 |
| LMArena Multi-Turn | 1177 | 1341 |
Frequently asked questions
Is Deepseek Coder v2 better than Step 3?
Step 3 is the stronger model overall, scoring 40.5 to 35.9 on the Noometry Index.
Is Deepseek Coder v2 or Step 3 better for coding?
Step 3 scores higher on coding benchmarks: 40.1 versus 38.1 in the Noometry coding category.
How many benchmarks do Deepseek Coder v2 and Step 3 share?
15 benchmarks have published results for both models. Deepseek Coder v2 has 24 scored results on Noometry and Step 3 has 17.