Key Specifications

Vendordeepseek
Versioncoder-33b
Release Date2024-01-25
Context Window16384 tokens
Input Modalitiestext
Output Modalitiestext
LicenseDeepSeek License
Documentationhttps://api-docs.deepseek.com/

Benchmark Performance

Benchmark Score Unit Evaluated At Notes Source
MMLU 65.9 % 2024-01-25 5-shot view
HUMANEVAL 71.8 pass@1 2024-01-25 view
GSM8K 69.7 % 2024-01-25 0-shot CoT view
MATH 32.1 % 2024-01-25 0-shot CoT view
BBH 61.6 % 2024-01-25 3-shot CoT view
GPQA 26.8 % 2024-01-25 0-shot view
IFEVAL 58.8 % 2024-01-25 prompt_strict view
ARC 92.2 % 2024-01-25 challenge view
MUSR 45.8 % 2024-01-25 0-shot view
WINOGRANDE 79.8 % 2024-01-25 0-shot view

Pricing

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

Source: https://api-docs.deepseek.com/quick_start/pricing · as of 2024-01-25

Compliance

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

DeepSeek Coder 33B

Model Overview

DeepSeek Coder 33B 代码专用开源模型, 16K 上下文, 33B 参数, 性能接近 GPT-3.5, 开源代码模型旗舰。

Core Specifications

Vendor Version Release Date Context Window Input Modalities Output Modalities License
Deepseek coder-33b 2024-01-25 16K text text DeepSeek License

Benchmark Performance

Benchmark Score Unit Notes
MMLU (Massive Multitask Language Understanding) 65.9 % 5-shot
HumanEval 71.8 pass@1
GSM8K (Grade School Math 8K) 69.7 % 0-shot CoT
MATH 32.1 % 0-shot CoT
BBH (BIG-Bench Hard) 61.6 % 3-shot CoT
GPQA 26.8 % 0-shot
IFEval 58.8 % prompt_strict
ARC 92.2 % challenge
MUSR 45.8 % 0-shot
WinoGrande 79.8 % 0-shot

Pricing

Input Output Cache Read Cache Write

per million tokens

Strengths

  • Reliable general-purpose model.

Weaknesses

  • Proprietary, not self-hostable.
  • Context window 16K is limited.

Use Cases

  • Code generation and debugging

References