LANGUAGE MODEL DeepSeek Last updated:

DeepSeek-Coder V2

Open Coding Frontier

DeepSeek's June 2024 coding model — 236B mixture-of-experts activating 21B per token, fine-tuned for code in 338 programming languages. At release, it beat Codestral 22B and matched closed- source GPT-4 Turbo on HumanEval. Open under a custom permissive license. The Coder V3 / V3.2 successors keep DeepSeek at the open coding frontier.

Intelligence
Below avg
Cost
Free
Open weights — self-host
Context
128K
Up to 128,000 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on Artificial Analysis. Cost buckets are fixed dollar thresholds keyed off output-token price ($/M out).
Intelligence
  • Top 1%≤ 1%
  • Top 5%≤ 5%
  • Top 10%≤ 10%
  • Good≤ 25%
  • Medium≤ 50%
  • Below avg> 50%
Speed
  • Top 1%≥ 345 tok/s
  • Top 5%≥ 237 tok/s
  • Top 10%≥ 196 tok/s
  • Good≥ 146 tok/s
  • Medium≥ 90 tok/s
  • Slow< 90 tok/s
Cost
  • Freeopen weights · self-host
  • Low< $1 / M out
  • Moderate$1–5 / M out
  • High≥ $5 / M out

Why it matters

Pushed open-weight coding models past GPT-4 Turbo. By late 2024, serious enterprise coding deployments started defaulting to DeepSeek-Coder for self-hosted Copilot replacements.

Core Capabilities

Generative
Produces images, video, audio, or other media.
Coding
Strong real-world software engineering.
Long Documents
Handles entire codebases, books, and multi-doc RAG.
Multimodal
Combines text, vision, and audio in one model.

Context Window

128k tokens
≈ 98 pages
4k Chat 聊天
32k Long docs 长文档
128k This model 本模型
400k Multi-doc 多文档
1M Codebase 整个代码库
10M

Availability

API
Not available
Product / App
Not available
Open Source
Released
Enterprise

Pricing Model

Free / self-host
Open weights — pay only for compute.
Self-host

Capability / Performance

Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).

Lower 20% Upper 80% This model
Reasoning
AA Intelligence Index · scaled to 10
1.7
5.6
1.5
Coding
SciCode · scaled to 10
1.8
4.3
1.4
Agentic tasks
Terminal-Bench Hard · scaled to 10
0.2
3.6
3.6
Context / memory
Context window size · log-scaled
6.0
9.0
6.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.

What it feels like

  • First open-weights coding model to beat GPT-4 Turbo, Claude 3 Opus, and Gemini 1.5 Pro on code + math
  • 236B MoE / 21B active — derived from DeepSeek-V2 with 6T extra tokens of code/math pre-training
  • HumanEval 90.2%, MBPP 76.2%, LiveCodeBench 43.4%, MATH 75.7% — frontier coding numbers at OSS price
  • Programming language support jumped from 86 to 338; context extended from 16K to 128K
  • 16B Lite variant (2.4B active) brings most of the quality to a single-GPU deployment
  • Set the playbook later refined into DeepSeek V3 / R1 / V4 — coding-specialist via continued pre-training

Best use cases

  • Self-hosted code-completion engines and IDE integrations
  • Repository-level QA and refactor pipelines using open weights
  • Multilingual code (338 languages) including niche legacy ones
  • Cost-sensitive bulk inference for code review / lint suggestions

Tools to try

Not ideal for

  • Frontier coding by mid-2025 — Claude Opus 4.5 / Sonnet 4.5 / DeepSeek V4 lead by wide margins
  • Edge / single-consumer-GPU deployments at the 236B scale (use Lite)
  • Non-coding chat — general dialogue is weaker than non-specialist DeepSeek models

Model Evolution

View full evolution tree →