DeepSeek Math 7B: Complete Benchmark Performance Guide

deepseek math-7b 2024-02-06 DeepSeek License

In-depth analysis of DeepSeek Math 7B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendordeepseek
Versionmath-7b
Release Date2024-02-06
Context Window4096 tokens
Input Modalitiestext
Output Modalitiestext
LicenseDeepSeek License
Documentationhttps://api-docs.deepseek.com/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 69.3 % 2024-02-06 5-shot view
HUMANEVAL 54 pass@1 2024-02-06 — view
GSM8K 83.1 % 2024-02-06 0-shot CoT view
MATH 53.8 % 2024-02-06 0-shot CoT view
BBH 65.9 % 2024-02-06 3-shot CoT view
GPQA 40.5 % 2024-02-06 0-shot view
IFEVAL 61.1 % 2024-02-06 prompt_strict view
ARC 90.5 % 2024-02-06 challenge view
MUSR 49.8 % 2024-02-06 0-shot view
WINOGRANDE 79.2 % 2024-02-06 0-shot view

Pricing

Tier Price Currency
Input$0.14 / MtokUSD
Output$0.14 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://api-docs.deepseek.com/quick_start/pricing · as of 2024-02-06

Compliance

  • Data Residency: CN
  • SOC2: ✗
  • HIPAA: ✗
  • GDPR: ✗
  • ISO 27001: ✗

DeepSeek Math 7B

Model Overview

DeepSeek Math 7B 数学专用开源模型, 4K 上下文, 在 MATH 基准上达到 64.7%, 7B 规模数学模型领先。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Deepseek math-7b 2024-02-06 4K text text DeepSeek License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 69.3 % 5-shot
HumanEval 54.0 pass@1 —
GSM8K (Grade School Math 8K) 83.1 % 0-shot CoT
MATH 53.8 % 0-shot CoT
BBH (BIG-Bench Hard) 65.9 % 3-shot CoT
GPQA 40.5 % 0-shot
IFEval 61.1 % prompt_strict
ARC 90.5 % challenge
MUSR 49.8 % 0-shot
WinoGrande 79.2 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 4K is limited.

Use Cases

  • General chat and Q&A

References