Model comparison
Codestral vs DeepSeek V4 Pro
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 30.6 on the Noometry Index. Codestral costs 2.2× less per token, which makes it the better buy when DeepSeek V4 Pro's lead doesn't matter for your workload.
Last verified . 2 shared benchmarks.
Mistral AI
DeepSeekSummary
- They share 2 benchmarks with published results for both. Codestral scores higher in 0 categories and DeepSeek V4 Pro in 2 categories; 2 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Pro leads 56.5 to 19.8.
- The biggest single-benchmark swing is Kagi LLM Benchmark: 32.5% for Codestral and 53.5% for DeepSeek V4 Pro.
- Codestral is cheaper at $0.30 / $0.90 per million input/output tokens, against $0.66 / $1.98 for DeepSeek V4 Pro.
- DeepSeek V4 Pro accepts more context: 1M tokens versus 256K.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| Codestral | DeepSeek V4 Pro | |
|---|---|---|
| Provider | Mistral AI | DeepSeek |
| Noometry Index | 30.6 | 54.3 |
| Released | 2024-05-29 | 2026-04-24 |
| Weights | Proprietary | Open |
| Context window | 256K | 1M |
| Max output | 8K | 393K |
| Input $ / M tokens | $0.30 | $0.66 |
| Output $ / M tokens | $0.90 | $1.98 |
| Results tracked | 7 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
Codestral: 27.3 (#321), DeepSeek V4 Pro: 52.4 (#34)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| ALE-Bench | 137.78 | 1,403 |
| SWE-bench Verified | — | 77.6% |
| FrontierCode | — | 28.6% |
| Aider Polyglot | 11.1% | — |
| LMArena WebDev | — | 1582 |
| SciCode | — | 51% |
| WeirdML | — | 66.2% |
| BigCodeBench Instruct | 41.8% | — |
| LMArena Coding | — | 1470 |
| BigCodeBench Complete | 52.5% | — |
| HumanEval+ | 73.8% | — |
| MBPP+ | 61.9% | — |
Agentic & Tool Use Not comparable
Codestral: —, DeepSeek V4 Pro: 32.8 (#58)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| APEX-Agents | — | 47.3% |
| Vending-Bench 2 | — | 3,285 |
Reasoning DeepSeek V4 Pro leads
Codestral: 19.8 (#251), DeepSeek V4 Pro: 56.5 (#24)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| Kagi LLM Benchmark | 32.5% | 53.5% |
| ARC-AGI-2 | — | 61.3% |
| NYT Connections (extended) | — | 91.3% |
| ARC-AGI-1 | — | 90.5% |
| CritPt | — | 18% |
| Chess Puzzles | — | 47% |
| LMArena Hard Prompts | — | 1461 |
| Mystery Game Puzzles | — | 43% |
| DTBench | — | 93.9% |
| LMCA | — | 45.5% |
| Surface Evolver Bench | — | 40% |
| Epoch Capabilities Index | — | 155.31 |
| ForecastBench | — | 56.1 |
Math Not comparable
Codestral: —, DeepSeek V4 Pro: 64.8 (#30)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| FrontierMath (Tiers 1-3) | — | 64.6% |
| FrontierMath Tier 4 | — | 26.8% |
| MathArena Final-Answer Competitions | — | 76.6% |
| OTIS Mock AIME 2024-2025 | — | 98.6% |
| ProofBench | — | 50% |
| LMArena Math | — | 1455 |
Knowledge Not comparable
Codestral: —, DeepSeek V4 Pro: 59.5 (#31)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| GPQA Diamond | — | 91.7% |
| SimpleQA Verified | — | 52.9% |
| Vectara Hallucination Rate | — | 8.6% |
| LMArena Expert | — | 1464 |
Multilingual Not comparable
Codestral: —, DeepSeek V4 Pro: 54.4 (#45)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | — | 1439 |
| LMArena Chinese | — | 1486 |
| LMArena French | — | 1472 |
| LMArena German | — | 1458 |
| LMArena Japanese | — | 1445 |
| LMArena Korean | — | 1447 |
| LMArena Russian | — | 1453 |
| LMArena Spanish | — | 1458 |
Instruction Following Not comparable
Codestral: —, DeepSeek V4 Pro: 76.1 (#47)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | — | 1448 |
Long Context Not comparable
Codestral: —, DeepSeek V4 Pro: 45.0 (#51)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| CL-bench Life | — | 13.5% |
| LMArena Longer Query | — | 1458 |
Writing & Preference Not comparable
Codestral: —, DeepSeek V4 Pro: 65.5 (#46)
| Benchmark | Codestral | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | — | 1451 |
| LMArena Creative Writing | — | 1446 |
| EQ-Bench Creative Writing | — | 1553 |
| EQ-Bench 4 | — | 1166 |
| LMArena Multi-Turn | — | 1467 |
Frequently asked questions
Is Codestral better than DeepSeek V4 Pro?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 30.6 on the Noometry Index. Codestral costs 2.2× less per token, which makes it the better buy when DeepSeek V4 Pro's lead doesn't matter for your workload.
Which is cheaper, Codestral or DeepSeek V4 Pro?
Codestral is cheaper. It lists at $0.30 per million input tokens and $0.90 per million output tokens; DeepSeek V4 Pro lists at $0.66 and $1.98.
Is Codestral or DeepSeek V4 Pro better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 27.3 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Pro does, with 1M tokens against 256K.
How many benchmarks do Codestral and DeepSeek V4 Pro share?
2 benchmarks have published results for both models. Codestral has 7 scored results on Noometry and DeepSeek V4 Pro has 48.