DeepSeek V2 Chat: Complete Benchmark Performance Guide

deepseek v2-chat 2024-06-17 DeepSeek License

In-depth analysis of DeepSeek V2 Chat performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendordeepseek
Versionv2-chat
Release Date2024-06-17
Context Window32768 tokens
Input Modalitiestext
Output Modalitiestext
LicenseDeepSeek License
Documentationhttps://api-docs.deepseek.com/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 76.6 % 2024-06-17 5-shot view
HUMANEVAL 67 pass@1 2024-06-17 — view
GSM8K 77.4 % 2024-06-17 0-shot CoT view
MATH 49.3 % 2024-06-17 0-shot CoT view
BBH 75.4 % 2024-06-17 3-shot CoT view
GPQA 38.3 % 2024-06-17 0-shot view
IFEVAL 71.5 % 2024-06-17 prompt_strict view
ARC 94.6 % 2024-06-17 challenge view
MUSR 52 % 2024-06-17 0-shot view
WINOGRANDE 80.2 % 2024-06-17 0-shot view

Pricing

Tier Price Currency
Input$0.14 / MtokUSD
Output$0.28 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://api-docs.deepseek.com/quick_start/pricing · as of 2024-06-17

Compliance

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

DeepSeek V2 Chat

Model Overview

DeepSeek V2 Chat 对话版本, 32K 上下文, 针对对话场景优化, 适合聊天助手与客服应用。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Deepseek v2-chat 2024-06-17 32K text text DeepSeek License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 76.6 % 5-shot
HumanEval 67.0 pass@1 —
GSM8K (Grade School Math 8K) 77.4 % 0-shot CoT
MATH 49.3 % 0-shot CoT
BBH (BIG-Bench Hard) 75.4 % 3-shot CoT
GPQA 38.3 % 0-shot
IFEval 71.5 % prompt_strict
ARC 94.6 % challenge
MUSR 52.0 % 0-shot
WinoGrande 80.2 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Mixture-of-Experts architecture.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • General chat and Q&A

References