DeepSeek LLM 67B: Complete Benchmark Performance Guide

deepseek llm-67b 2024-01-05 DeepSeek License

In-depth analysis of DeepSeek LLM 67B performance on MMLU, HumanEval, GSM8K, MATH, BBH benchmarks, including pricing and use cases.

Key Specifications

Vendordeepseek
Versionllm-67b
Release Date2024-01-05
Context Window32768 tokens
Input Modalitiestext
Output Modalitiestext
LicenseDeepSeek License
Documentationhttps://api-docs.deepseek.com/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 75.2 % 2024-01-05 5-shot view
HUMANEVAL 68.3 pass@1 2024-01-05 — view
GSM8K 79.9 % 2024-01-05 0-shot CoT view
MATH 39.3 % 2024-01-05 0-shot CoT view
BBH 72.6 % 2024-01-05 3-shot CoT view
GPQA 37.3 % 2024-01-05 0-shot view
IFEVAL 76.8 % 2024-01-05 prompt_strict view
ARC 94 % 2024-01-05 challenge view
MUSR 51.9 % 2024-01-05 0-shot view
WINOGRANDE 78.1 % 2024-01-05 0-shot view

Pricing

Tier Price Currency
Input$0.4 / MtokUSD
Output$0.4 / MtokUSD
Cache Read$0 / MtokUSD
Cache Write$0 / MtokUSD

Source: https://api-docs.deepseek.com/quick_start/pricing · as of 2024-01-05

Compliance

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

DeepSeek LLM 67B

Model Overview

DeepSeek LLM 67B 通用开源模型, 32K 上下文, 中英文能力平衡, 在开源 67B 模型中表现突出。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Deepseek llm-67b 2024-01-05 32K text text DeepSeek License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 75.2 % 5-shot
HumanEval 68.3 pass@1 —
GSM8K (Grade School Math 8K) 79.9 % 0-shot CoT
MATH 39.3 % 0-shot CoT
BBH (BIG-Bench Hard) 72.6 % 3-shot CoT
GPQA 37.3 % 0-shot
IFEval 76.8 % prompt_strict
ARC 94.0 % challenge
MUSR 51.9 % 0-shot
WinoGrande 78.1 % 0-shot

Pricing

Input Output Cache Read Cache Write
— — — —

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.

Use Cases

  • General chat and Q&A

References