Codestral Mamba: Complete Benchmark Performance Guide

mistral codestral-mamba 2024-07-16 Apache 2.0

In-depth analysis of Codestral Mamba performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendormistral
Versioncodestral-mamba
Release Date2024-07-16
Context Window256000 tokens
Input Modalitiestext
Output Modalitiestext
LicenseApache 2.0
Documentationhttps://docs.mistral.ai/

Benchmark Performance

BenchmarkScoreUnitEvaluated AtNotesSource
MMLU59.3%2024-07-165-shotview
HUMANEVAL86.2pass@12024-07-16—view
GSM8K67.8%2024-07-160-shot CoTview
MATH28.1%2024-07-160-shot CoTview
BBH66%2024-07-163-shot CoTview
GPQA36.3%2024-07-160-shotview
IFEVAL65.5%2024-07-16prompt_strictview
ARC87.4%2024-07-16challengeview
MUSR47%2024-07-160-shotview
WINOGRANDE74.1%2024-07-160-shotview

Pricing

TierPriceCurrency
Input$0.25 / MtokUSD
Output$0.25 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://mistral.ai/technology/ · as of 2024-07-16

Compliance

  • Data Residency: EU
  • SOC2: ✓
  • HIPAA: ✗
  • GDPR: ✓
  • ISO 27001: ✓

Codestral Mamba

Model Overview

Mistral Codestral Mamba 7B 代码模型, 256K 上下文, 基于 Mamba 架构, 线性时间复杂度适合长序列。

Core Specifications

VendorVersionRelease DateContext WindowInput ModalitiesOutput ModalitiesLicense
Mistralcodestral-mamba2024-07-16256KtexttextApache 2.0

Benchmark Performance

BenchmarkScoreUnitNotes
MMLU (Massive Multitask Language Understanding)59.3%5-shot
HumanEval86.2pass@1—
GSM8K (Grade School Math 8K)67.8%0-shot CoT
MATH28.1%0-shot CoT
BBH (BIG-Bench Hard)66.0%3-shot CoT
GPQA36.3%0-shot
IFEval65.5%prompt_strict
ARC87.4%challenge
MUSR47.0%0-shot
WinoGrande74.1%0-shot

Pricing

InputOutputCache ReadCache Write
————

per million tokens

Strengths

  • HumanEval 86.2, excellent code generation.
  • Context window 256K.

Weaknesses

  • MMLU 59.3, weak knowledge reasoning.
  • Proprietary, not self-hostable.

Use Cases

  • Code generation and debugging
  • Long document summarization

References