Model comparison
Codestral 25.01 vs Qwen2.5-Coder
- Input
- $0.30 / 1M tokens
- Output
- $0.90 / 1M tokens
- Context
- 256K
- Released
- 2025-01-13
- Input
- $0.66 / 1M tokens
- Output
- $1.00 / 1M tokens
- Context
- 128K (131,072 tokens) for 7B/14B/32B; 32K for 0.5B/1.5B/3B
- Released
- 2024-11-12
Qwen2.5-Coder scores higher on 2 of the 3 benchmarks both models report (the other leads on 1); Codestral 25.01 is cheaper on input tokens ($0.3 vs $0.7 per 1M, 2.2×); Codestral 25.01 has the larger context window (256K vs 128K); Qwen2.5-Coder ships open weights while Codestral 25.01 is API-only.
Specs and pricing side by side
| Spec | Codestral 25.01 | Qwen2.5-Coder |
|---|---|---|
| Maker | Mistral AI | Alibaba (Qwen) |
| Released | 2025-01-13 | 2024-11-12 |
| Context window | 256K | 128K (131,072 tokens) for 7B/14B/32B; 32K for 0.5B/1.5B/3B |
| Max output | Not disclosed | 8,192 tokens (default generation); hosted… |
| Input price | $0.30 / 1M tokens | $0.66 / 1M tokens |
| Output price | $0.90 / 1M tokens | $1.00 / 1M tokens |
| Parameters | Not publicly disclosed | Family of 6 sizes: 0.5B, 1.5B, 3B, 7B, 14B, 32B (flagship 32B = 32.5B total / 31.0B non-embedding) |
| License | Proprietary (Mistral commercial / API-only — not released as open weights, unlike the original Codestral 22B) | Apache 2.0 (all sizes except the 3B model, which uses the Qwen Research license) |
| Weights | API only | Open weights |
| Modalities | Text | Textsame |
| Knowledge cutoff | Not disclosed | Not officially published for Qwen2.5-Coder;… |
Benchmarks both models report
| Benchmark | Codestral 25.01 | Qwen2.5-Coder |
|---|---|---|
| HumanEval | 86.6% | 92.7% |
| MBPP | 80.2% | 90.2% |
| LiveCodeBench | 37.9% | 31.4% |
Scores are as published by each maker or its cited source; harness, reasoning effort and date can differ between rows, so treat small gaps as ties. Sources are listed on each model page.
Which one should you pick?
Pick Codestral 25.01 if…
- You care about LiveCodeBench (it scores higher on 1 shared benchmark).
- Cost matters: input tokens cost 2.2× less ($0.3 vs $0.7 per 1M).
- You need the longer context (256K vs 128K).
- Your use case is: real-time IDE code completion and fill-in-the-middle inside editors like VS Code and JetBrains (via the Continue plugin).
Pick Qwen2.5-Coder if…
- You care about HumanEval, MBPP (it scores higher on 2 shared benchmarks).
- You want to self-host or fine-tune (open weights, Apache 2.0 (all sizes except the 3B model, which uses the Qwen Research license)).
- Your use case is: self-hosted code assistants and chat-based coding helpers where open weights matter.
FAQ
Which is better, Codestral 25.01 or Qwen2.5-Coder?
Qwen2.5-Coder scores higher on 2 of the 3 benchmarks both models report (the other leads on 1); Codestral 25.01 is cheaper on input tokens ($0.3 vs $0.7 per 1M, 2.2×); Codestral 25.01 has the larger context window (256K vs 128K); Qwen2.5-Coder ships open weights while Codestral 25.01 is API-only. The better pick depends on whether you value benchmark scores, price per token, context length or openness for your workload.
How much cheaper is Codestral 25.01?
Codestral 25.01: $0.30 / 1M tokens input, $0.90 / 1M tokens output. Qwen2.5-Coder: $0.66 / 1M tokens input, $1.00 / 1M tokens output (list API prices as published by each maker).
What are the context windows of Codestral 25.01 and Qwen2.5-Coder?
Codestral 25.01 offers 256K with up to Not disclosed output; Qwen2.5-Coder offers 128K (131,072 tokens) for 7B/14B/32B; 32K for 0.5B/1.5B/3B with up to 8,192 tokens (default generation); hosted endpoints commonly cap output at 32,768 output.
Is Codestral 25.01 newer than Qwen2.5-Coder?
Codestral 25.01 was released 2025-01-13; Qwen2.5-Coder was released 2024-11-12.