Codestral Base: pricing, context window and benchmarks
Codestral Base is a open-weight model from Mistral AI. It offers a 8,192-token context window, costs $3.31 per million input tokens and $16.95 per million output tokens, and averages 73.1 across the index's benchmark suite.
Pricing
Codestral Base costs $3.31 per million input tokens and $16.95 per million output tokens. A blended 3:1 input-to-output mix works out to about $6.72 per million tokens. Cached input is listed at $0.50 per million tokens.
| Item | Price | Notes |
|---|---|---|
| Input | $3.31 | per 1M tokens |
| Output | $16.95 | per 1M tokens |
| Cached input | $0.50 | per 1M tokens |
| Blended 3:1 | $6.72 | estimate per 1M tokens |
| 1M in + 1M out | $20.26 | estimate |
Benchmarks
Scores below are modelled estimates on a 0–100 scale. Use them to compare relative position, not as official vendor results.
| Benchmark | Score | Rank | Percentile | Relative |
|---|---|---|---|---|
| MMLU-Pro | 74.1 | 348 | 42th | |
| GPQA Diamond | 70.4 | 252 | 58th | |
| SWE-bench Verified | 62.6 | 264 | 56th | |
| HumanEval | 79.7 | 240 | 60th | |
| MATH-500 | 77.4 | 267 | 56th | |
| AIME 2025 | 68.7 | 283 | 53th | |
| LiveCodeBench | 65.7 | 279 | 54th | |
| IFEval | 79.2 | 347 | 42th | |
| TruthfulQA | 74.3 | 347 | 42th | |
| HellaSwag | 83.3 | 311 | 48th | |
| MMMU | 68.3 | 339 | 44th | |
| Arena Elo | 1371.7 | 310 | 48th |
What Codestral Base is good for
Codestral Base is an open-weight model, so it can be self-hosted or run through a third-party inference provider. That matters for data residency, fine-tuning and long-run cost control. At $3.31 per million input tokens (hosted list price) and a 8,192-token context window, it is a practical option for teams that want more control than a closed API allows.
Strengths
- Input price of $3.31 per million tokens is at or below the index median.
- Open weights make self-hosting, fine-tuning and data-residency review possible.
- Accepts more than text, including image input where the provider supports it.
Trade-offs
- Context window of 8,192 tokens is small for long-document work.
How it compares
| Model | Provider | Context | Input / 1M | Avg score |
|---|---|---|---|---|
| Gemini 3 Pro Nano | Google DeepMind | 8192 | $9.11 | 73.2 |
| Codestral Base | Mistral AI | 8192 | $3.31 | 73.1 |
| Mixtral Groq | Groq | 16384 | $3.37 | 73.1 |
| Claude 3.7 Sonnet Mini | Anthropic | 32768 | $9.74 | 73.0 |
| Llama 3.1 Turbo Base | Together AI | 16384 | $3.30 | 73.0 |
| Nemotron 4 Base | NVIDIA | 8192 | $3.29 | 73.0 |
FAQ
How much does Codestral Base cost?
Codestral Base is listed at $3.31 per million input tokens and $16.95 per million output tokens. Cached input is $0.50 per million tokens. A blended 3:1 input-to-output mix is about $6.72 per million tokens.
What is the context window of Codestral Base?
Codestral Base has a context window of 8,192 tokens and can emit up to 8,192 output tokens per request in this index.
Is Codestral Base open weights?
Yes. Codestral Base is listed as an open-weight model under a open weights licence.
How does Codestral Base score on benchmarks?
Codestral Base averages 73.1 across the scored benchmarks in this index, ranking #299 of 600 models. Its strongest relative result is Arena Elo at 1371.7.
What is Codestral Base best for?
Codestral Base is an open-weight model, so it can be self-hosted or run through a third-party inference provider. That matters for data residency, fine-tuning and long-run cost control. At $3.31 per million input tokens (hosted list price) and a 8,192-token context window, it is a practical option for teams that want more control than a closed API allows.