Codestral vs DeepSeek-V3: Detailed Comparison
Choosing between Codestral (Mistral AI) and
DeepSeek-V3 (DeepSeek) comes down to three things:
per-token pricing, context window, and which capability matters most for your workload.
Codestral costs $0.20/M input vs
$0.27/M for DeepSeek-V3;
context windows are 32K vs
128K tokens. Detailed breakdown below.
Side-by-side specs
| Spec | Codestral | DeepSeek-V3 |
|---|
| Provider | Mistral AI | DeepSeek |
|---|
| Released | 2024-05-29 | 2024-12-26 |
|---|
| Input price |
$0.20/M |
$0.27/M |
| Output price |
$0.60/M |
$1.10/M |
| Cached input |
— |
$0.0700/M |
| Context window |
32K |
128K |
| Max output |
4K |
8K |
| Modalities |
text |
text |
| Tokenizer |
mistral |
deepseek |
Capability matrix
| Capability | Codestral | DeepSeek-V3 |
|---|
| function calling |
Yes |
Yes |
| json mode |
Yes |
Yes |
| streaming |
Yes |
Yes |
| code |
Yes |
No |
Benchmark comparison
Higher is better for all benchmarks shown.
| Benchmark | Category | Codestral | DeepSeek-V3 | Δ |
|---|
| MMLU-Pro |
general |
— |
75.9 |
— |
| HumanEval |
coding |
— |
82.6 |
— |
Per-call cost on typical workloads
| Workload (in/out tokens) | Codestral | DeepSeek-V3 | Cheaper by |
|---|
| Standard chat (1K / 500) |
$0.000500 |
$0.000820 |
Codestral by $0.000320 |
| RAG (4K / 500) |
$0.001100 |
$0.001630 |
Codestral by $0.000530 |
| Long doc (20K / 1K) |
$0.004600 |
$0.006500 |
Codestral by $0.001900 |
| Very long context (100K / 2K) |
$0.020900 |
$0.028650 |
Codestral by $0.007750 |
When to choose Codestral over DeepSeek-V3
- Per-token input cost is 26% lower — meaningful for high-volume workloads.
- Supports code — DeepSeek-V3 does not.
When to choose DeepSeek-V3 over Codestral
- Larger context window (128K vs 32K).
Related comparisons