codestral-2508: benchmark results
Mistral's Codestral 2508, a code-specialized model for fill-in-the-middle completion, code generation, correction and test generation, with a 256K-token context window. Provider: Mistral. Released 2025-07-30. Access: API.
Aggregate Rating 1572 ± 22, rank #537 of 1635 rated models, from 28 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Tinybird AI SQL Benchmark - First-Attempt Success Rate | 100 | Questions answered with a valid query on the first attempt ( | 94.2 |
| Tinybird AI SQL Benchmark - Success Rate | 100 | Questions answered with a valid query within 3 retries (%) | 71.9 |
| Guesswork 2026-07 | 1 | MAE (z-score units) | 66.7 |
| Intent2Tx - State Equivalence | 48.1 | State equivalence (from 0-1, times 100): share of generated | 50 |
| Tinybird AI SQL Benchmark - Exactness | 47.22 | Result exactness vs human reference queries (0-100) | 42.3 |
| Guesswork 2026-08 | 1.21 | MAE (z-score units) | 40.6 |
| Intent2Tx (Retrieval, Multi-Step) | 52 | Final weighted score (from 0-1, times 100): per-instance 0.1 | 40 |
| Wolfram LLM Benchmarking Project | 37.8 | Correct Functionality (%) | 38.3 |
| Guesswork 2026-10 | 1.14 | MAE (z-score units) | 37.5 |
| AI Chess Leaderboard (Reasoning) | 627 | Elo | 36.6 |
| Guesswork 2026-09 | 0.92 | MAE (z-score units) | 36.1 |
| Intent2Tx | 31 | Final weighted score (from 0-1, times 100): per-instance 0.1 | 35.7 |
Interactive version: theaggregate.ai/model?slug=codestral-2508 · How It Works · Data refreshed daily, snapshot 2026-11-05.