Model comparison
Codestral vs Deepseek Coder v2
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 30.6 on the Noometry Index.
Last verified . 4 shared benchmarks.
Mistral AI
DeepSeekSummary
- They share 4 benchmarks with published results for both. Codestral scores higher in 0 categories and Deepseek Coder v2 in 2 categories; 2 gaps are clear of the uncertainty.
- The widest gap is in coding, where Deepseek Coder v2 leads 38.1 to 27.3.
- The biggest single-benchmark swing is BigCodeBench Complete: 52.5% for Codestral and 59.7% for Deepseek Coder v2.
- Deepseek Coder v2 has downloadable open weights; the other is API-only.
Side by side
| Codestral | Deepseek Coder v2 | |
|---|---|---|
| Provider | Mistral AI | DeepSeek |
| Noometry Index | 30.6 | 35.9 |
| Released | 2024-05-29 | 2024-06-17 |
| Weights | Proprietary | Open |
| Context window | 256K | — |
| Max output | 8K | — |
| Input $ / M tokens | $0.30 | — |
| Output $ / M tokens | $0.90 | — |
| Results tracked | 7 | 24 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Deepseek Coder v2 leads
Codestral: 27.3 (#321), Deepseek Coder v2: 38.1 (#183)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| BigCodeBench Instruct | 41.8% | 48.2% |
| BigCodeBench Complete | 52.5% | 59.7% |
| HumanEval+ | 73.8% | 82.3% |
| MBPP+ | 61.9% | 75.1% |
| Aider Polyglot | 11.1% | — |
| LMArena Coding | — | 1251 |
| ALE-Bench | 137.78 | — |
Reasoning Deepseek Coder v2 leads
Codestral: 19.8 (#251), Deepseek Coder v2: 23.6 (#176)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| Kagi LLM Benchmark | 32.5% | — |
| LMArena Hard Prompts | — | 1207 |
| WinoGrande | — | 83.7% |
Math Not comparable
Codestral: —, Deepseek Coder v2: 34.9 (#190)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Math | — | 1241 |
| GSM8K | — | 94.5% |
Knowledge Not comparable
Codestral: —, Deepseek Coder v2: 32.3 (#212)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Expert | — | 1181 |
| ARC (AI2) Challenge | — | 64.3% |
Multilingual Not comparable
Codestral: —, Deepseek Coder v2: 36.3 (#240)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Non-English | — | 1182 |
| LMArena Chinese | — | 1201 |
| LMArena French | — | 1185 |
| LMArena German | — | 1164 |
| LMArena Japanese | — | 1126 |
| LMArena Korean | — | 1104 |
| LMArena Russian | — | 1188 |
| LMArena Spanish | — | 1153 |
Instruction Following Not comparable
Codestral: —, Deepseek Coder v2: 61.7 (#242)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Instruction Following | — | 1180 |
Long Context Not comparable
Codestral: —, Deepseek Coder v2: 37.0 (#224)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Longer Query | — | 1219 |
Writing & Preference Not comparable
Codestral: —, Deepseek Coder v2: 38.2 (#253)
| Benchmark | Codestral | Deepseek Coder v2 |
|---|---|---|
| LMArena Text | — | 1191 |
| LMArena Creative Writing | — | 1120 |
| LMArena Multi-Turn | — | 1177 |
Frequently asked questions
Is Codestral better than Deepseek Coder v2?
Deepseek Coder v2 is the stronger model overall, scoring 35.9 to 30.6 on the Noometry Index.
Is Codestral or Deepseek Coder v2 better for coding?
Deepseek Coder v2 scores higher on coding benchmarks: 38.1 versus 27.3 in the Noometry coding category.
How many benchmarks do Codestral and Deepseek Coder v2 share?
4 benchmarks have published results for both models. Codestral has 7 scored results on Noometry and Deepseek Coder v2 has 24.