Model
Model explorer

DeepSeek Coder 33B

OPEN
DeepSeek · DeepSeek Coder family · released Nov 2, 2023

DeepSeek's first public model; open code LLM trained 87% code, beat Codex/GPT-3.5-turbo on HumanEval.

ReasoningCodingVisionFunction callingTool useAgentic
833.4
Elo · unrated
Parameters
33.3B
Active params
33.3B (dense)
Context
16K tokens
Architecture
Dense transformer decoder, code-specialized, 16K context
License
DeepSeek Model License (custom, commercial use permitted)
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
HumanEvalCoding79.3%#57
best: Claude Opus 4.5 · 99.4%
MBPPCoding70.0%#38
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
RepoBenchCoding41.9%#1
best: this model · 41.9%
Run it locally
VRAM @ Q4
20 GB
VRAM @ FP16
67 GB
Fits on (Q4)
RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Q4 (~20GB) fits within a 24GB RTX 4090, but no verified specific tokens/sec benchmark was found for this exact configuration.
Quantizations
GGUF · AWQ · GPTQ
DeepSeek Coder family
Elo progression across releases
DeepSeek Coder 33BNov 2023833.4
Resources
HFdeepseek-ai/deepseek-coder-33b-instructGHdeepseek-ai/DeepSeek-CoderDOCSdeepseekcoder.github.ioAPIAvailable via DeepSeek API
API price weights · each benchmark row carries its own source badge (see methodology)