Model Cards / Mistral AI

Codestral Release

model card1,242 words·5 min read·Aug 20, 2026·Source
Version History
Chaptered summary is still being generated for this document. Showing a heuristic brief in the meantime.
Summary
1,242-word document condensed to 68 words. Mistral AI · Aug 20, 2026
TL;DR

We introduce Codestral, our first-ever code model. Codestral is an open-weight generative AI model explicitly designed for code generation tasks.

Top benchmarks
BenchmarkVariantScore
HumanEvalinstruction-tuned, kotlin, pass_rate73.8%
HumanEvalkotlin, pass_rate72.0%
HumanEvalkotlin, pass_rate54.7%

Showing top 3 of 11. See full list below.

Capability claim
  • We introduce Codestral, our first-ever code model.
Deployment scope
  • available on demand by [reaching out to the team](/contact/).

Every italicized passage is a verbatim substring of the source document (checked deterministically after extraction). Field selection is heuristic — some quotes may lack surrounding context and some claims may be absent if no matching pattern appeared. For citation, open the source: original model card · source SHA 30b066c41790 · version dated Aug 20, 2026.

Extracted Evaluations(11 results)

Sort by:0/11 rows fully reproducible (0%)
BenchmarkCategoryStateScoreSetupSource
/ kotlin
codingscored
73.8%
pass rate
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
/ kotlin
codingscored
72.0%
pass rate
missing: shot countmissing: methodmissing: languagemissing: training state
self-reported
/ kotlin
codingscored
54.7%
pass rate
missing: shot countmissing: methodmissing: languagemissing: training state
self-reported
/ fim
codingmentioned
pass at 1
missing: shot countmissing: methodmissing: languagemissing: training state
self-reported
/ additional_languages
codingmentioned
pass at 1
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
/ fim
codingmentioned
pass at 1
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
codingmentioned
pass at 1
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
/ sanitized
codingmentioned
pass at 1
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
CruxEval
othermentioned
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
RepoBench
othermentioned
exact match
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported
Spider
othermentioned
instruction-tunedmissing: shot countmissing: methodmissing: language
self-reported