Model comparison

Codestral vs Mixtral 8x7B

Codestral is the stronger model overall, scoring 30.6 to 27.1 on the Noometry Index.

Last verified . 2 shared benchmarks.

Codestral Mistral AI

30.6

Rank #290 Reported

Mixtral 8x7B Mistral AI

27.1

Rank #334 Confirmed

Summary

  • They share 2 benchmarks with published results for both. Codestral scores higher in 1 category and Mixtral 8x7B in 1 category; 2 gaps are clear of the uncertainty.
  • The widest gap is in coding, where Mixtral 8x7B leads 32.8 to 27.3.
  • Codestral is cheaper at $0.30 / $0.90 per million input/output tokens, against $0.70 / $0.70 for Mixtral 8x7B.
  • Codestral accepts more context: 256K tokens versus 32K.
  • Mixtral 8x7B has downloadable open weights; the other is API-only.

Side by side

Codestral and Mixtral 8x7B specifications
CodestralMixtral 8x7B
ProviderMistral AIMistral AI
Noometry Index30.627.1
Released2024-05-292023-12-11
WeightsProprietaryOpen
Context window256K32K
Max output8K32K
Input $ / M tokens$0.30$0.70
Output $ / M tokens$0.90$0.70
Results tracked738

Sponsored placements are available on pages like this one. Advertise on Noometry

Category by category

Coding Mixtral 8x7B leads

Codestral: 27.3 (#321), Mixtral 8x7B: 32.8 (#269)

Coding benchmarks
BenchmarkCodestralMixtral 8x7B
HumanEval+73.8%39.6%
MBPP+61.9%49.7%
Aider Polyglot11.1%—
BigCodeBench Instruct41.8%—
LMArena Coding—1126
BigCodeBench Complete52.5%—
ALE-Bench137.78—

Reasoning Codestral leads

Codestral: 19.8 (#251), Mixtral 8x7B: 18.2 (#285)

Reasoning benchmarks
BenchmarkCodestralMixtral 8x7B
Kagi LLM Benchmark32.5%—
LMArena Hard Prompts—1115
DTBench—49.6%
Adversarial NLI—55.2%
Epoch Capabilities Index—118.47
ForecastBench—56.3
HellaSwag—86.7%
PIQA—83.6%
WinoGrande—77.2%

Math Not comparable

Codestral: —, Mixtral 8x7B: 18.8 (#289)

Math benchmarks
BenchmarkCodestralMixtral 8x7B
Omni-MATH—10.5%
LMArena Math—1147
MATH Level 5—10%
GSM8K—74.4%

Knowledge Not comparable

Codestral: —, Mixtral 8x7B: 11.0 (#301)

Knowledge benchmarks
BenchmarkCodestralMixtral 8x7B
GPQA Diamond—30.6%
MMLU-Pro—33.5%
GPQA (HELM)—29.6%
LMArena Expert—1088
ARC (AI2) Challenge—87.3%
MMLU—70.6%
OpenBookQA—85.8%
TriviaQA—82.2%

Multilingual Not comparable

Codestral: —, Mixtral 8x7B: 29.6 (#266)

Multilingual benchmarks
BenchmarkCodestralMixtral 8x7B
LMArena Non-English—1077
LMArena Chinese—1055
LMArena French—1166
LMArena German—1114
LMArena Japanese—931
LMArena Korean—968
LMArena Russian—1090
LMArena Spanish—1111

Instruction Following Not comparable

Codestral: —, Mixtral 8x7B: 51.0 (#297)

Instruction Following benchmarks
BenchmarkCodestralMixtral 8x7B
IFEval—57.5%
LMArena Instruction Following—1109

Long Context Not comparable

Codestral: —, Mixtral 8x7B: 33.4 (#260)

Long Context benchmarks
BenchmarkCodestralMixtral 8x7B
LMArena Longer Query—1103

Writing & Preference Not comparable

Codestral: —, Mixtral 8x7B: 34.2 (#270)

Writing & Preference benchmarks
BenchmarkCodestralMixtral 8x7B
LMArena Text—1132
LMArena Creative Writing—1109
WildBench—67.3%
LMArena Multi-Turn—1115

Frequently asked questions

Is Codestral better than Mixtral 8x7B?

Codestral is the stronger model overall, scoring 30.6 to 27.1 on the Noometry Index.

Which is cheaper, Codestral or Mixtral 8x7B?

Codestral is cheaper. It lists at $0.30 per million input tokens and $0.90 per million output tokens; Mixtral 8x7B lists at $0.70 and $0.70.

Is Codestral or Mixtral 8x7B better for coding?

Mixtral 8x7B scores higher on coding benchmarks: 32.8 versus 27.3 in the Noometry coding category.

Which has the bigger context window?

Codestral does, with 256K tokens against 32K.

How many benchmarks do Codestral and Mixtral 8x7B share?

2 benchmarks have published results for both models. Codestral has 7 scored results on Noometry and Mixtral 8x7B has 38.

Related comparisons

Go deeper