DeepSeek-OCR 2
DeepSeek-OCR 2 is DeepSeek's 3B-parameter document understanding model based on Visual Causal Flow. It converts document images into structured text or Markdown and supports dynamic-resolution OCR, layout grounding, and local serving through Transformers, vLLM, and SGLang.
Task Fit
Not marked for typed decisions in the current library.
Not marked for code agent in the current library.
Not marked for code in the current library.
Not marked for chat in the current library.
Document QA benefits from long context and instruction following.
Image or visual understanding, not necessarily image generation.
Not marked for image generation in the current library.
Not marked for video generation in the current library.
Not marked for voice in the current library.
Evidence level
How this is assessed →Memory and hardware-fit values are planning estimates unless a tested report is linked.
Variants and Quant Artifacts
Choose the artifact first; hardware fit follows from RAM, VRAM, format, and runtime.
| Quant | Format | Quality | Min RAM | Reco RAM | Runtime | Action |
|---|---|---|---|---|---|---|
| BF16 | safetensors | high | 12GB | 16GB | transformers, vllm, sglang, docker-model-runner | Plan with this |
Recommended Hardware
Lowest estimated 5-year cost that can run this model.
Enough unified/system memory with a balanced 5-year cost.
Highest local performance signal among compatible hardware.
Benchmarks
No benchmark data is available for this model yet.
Sources and compatibility
Execution evidence
Run DeepSeek-OCR 2 with a documented recipe
Recipes connect hardware, a model artifact, tools, settings, verification, and a reportable result.
No verified recipe is linked to this record yet.
Compatibility estimates remain available in the planner. A recipe appears here only after its exact stack and verification protocol are documented.
Similar Models
DeepSeek V4.1 Flash is DeepSeek's MIT-licensed open-weight multimodal MoE for reasoning, coding, agents, and native image understanding. The complete checkpoint is approximately 763B parameters, including a 552B backbone; its Causal Encoder-Decoder architecture activates 8B parameters during input prefill and 16B during output decoding. CSA2 and FP4 KV caching reduce long-context cache cost. The model supports a 1M-token context window and is available through official FP8 weights, the DeepSeek API, and Ollama Cloud. It remains a data-center-class deployment despite the low active-parameter count.
DeepSeek V4.1 Flash 1017 is an updated MIT-licensed 763B-parameter multimodal MoE checkpoint. It keeps the V4.1 family's 1M-token context, coding, reasoning, tool use, and image understanding, while remaining a data-center-class deployment despite activating far fewer parameters per token.
DeepSeek-VL2 is the full 27.5B-parameter MoE vision-language model, activating about 4.5B parameters per token. It provides the strongest capability in the VL2 family for visual question answering, OCR, document, table and chart understanding, and visual grounding.