# Janus-Pro

Janus-Pro is a unified multimodal understanding-and-generation model released by DeepSeek on January 27, 2025, in 1B and 7B sizes, which both interprets images and text and generates images from text prompts within a single autoregressive model.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[2](https://github.com/deepseek-ai/Janus)</sup> It is an advanced version of the earlier Janus model, improving both multimodal understanding and visual generation.<sup>[2](https://github.com/deepseek-ai/Janus)</sup>

| Key fact | Detail |
|---|---|
| Release date | January 27, 2025<sup>[2](https://github.com/deepseek-ai/Janus)</sup> |
| Sizes | Janus-Pro-1B and Janus-Pro-7B, built on DeepSeek-LLM 1.5B and 7B bases<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup> |
| Architecture | Decoupled visual encoding: SigLIP-L for understanding, VQ tokenizer for generation, one autoregressive transformer<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup> |
| Headline scores (vendor-reported) | GenEval 0.80, DPG-Bench 84.19, MMBench 79.2 for the 7B model<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> |
| Training compute | About 9 days (1.5B) and 14 days (7B) on 128 and 256 Nvidia A100 40GB GPUs<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> |
| Licensing | MIT-licensed code; model weights under the DeepSeek Model License, which permits commercial use under its terms<sup>[2](https://github.com/deepseek-ai/Janus)</sup> |
| Native resolution | 384 × 384 for both understanding and generation<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup> |

## What Janus-Pro is

Janus-Pro is a unified multimodal large language model (MLLM): one model that accepts images and text as input and also produces images as output.<sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup> This differs from a standard vision-language model that only captions or answers questions about images, and from a diffusion model such as [Stable Diffusion](https://www.edgepedia.org/stable-diffusion) that only generates images. In Janus-Pro both capabilities run through a single autoregressive transformer, meaning the model generates images token by token in the same way it generates text.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>

The model has a 4096-token context length and a 100K vocabulary, and comes in two sizes: 1B, built on DeepSeek-LLM-1.5b-base, and 7B, built on DeepSeek-LLM-7b-base.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup>

## Architecture: decoupled visual encoding

The core design principle is to decouple visual encoding for understanding and generation.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> [Understanding](https://www.edgepedia.org/understanding) uses SigLIP-Large-Patch16-384 as the vision encoder, supporting 384 × 384 image input; generation uses a VQ tokenizer with a codebook of 16,384 entries and a downsample rate of 16.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[3](https://huggingface.co/deepseek-ai/Janus-Pro-1B)</sup> Both pathways feed the same autoregressive transformer.

The predecessor Janus, released in October 2024, introduced this design and argued that a single shared visual encoder causes task conflict, because understanding needs semantic representations while generation needs fine-grained detail. Ablation experiments in the Janus paper showed that a shared encoder sacrificed understanding capability.<sup>[4](https://arxiv.org/html/2410.13848)</sup>

## Training and release as published

Training followed a three-stage recipe with understanding-to-generation-to-mixed data ratios of 1:0:3, 2:3:5, and 5:1:4 in the three stages; the Stage III ratio changed from Janus's 7:3:10. All images were trained at 384 × 384. Compared with Janus, Janus-Pro added roughly 90 million understanding samples and about 72 million synthetic aesthetic samples, with a 1:1 real-to-synthetic ratio in Stage II, and dropped ImageNet from Stage II.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>

The whole training process took about 9 days for the 1.5B model on 16 nodes and 14 days for the 7B model on 32 nodes, each node equipped with 8 Nvidia A100 (40GB) GPUs, that is 128 and 256 A100s respectively, using DeepSeek's HAI-LLM framework.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> For comparison, Janus trained in 7 days on a 16-node A100 cluster, so Janus-Pro represents a larger data and model scale.<sup>[4](https://arxiv.org/html/2410.13848)</sup>

The code repository is licensed under the [MIT License](https://www.edgepedia.org/mit-license), while use of the Janus models is subject to the DeepSeek Model License, which permits commercial usage under its terms. Weights are downloadable from [Hugging Face](https://www.edgepedia.org/hugging-face), and the family comprises Janus-1.3B, JanusFlow-1.3B, Janus-Pro-1B, and Janus-Pro-7B, all with 4096 sequence length, with an online demo available.<sup>[2](https://github.com/deepseek-ai/Janus)</sup>

## Benchmark results: vendor versus independent

All headline numbers below are vendor-reported by DeepSeek in the Janus-Pro paper; the sources note that they were internal tests by DeepSeek and that independent verification was still pending.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup>

- **Image generation (GenEval):** Janus-Pro-7B scores 0.80, which DeepSeek reports as outperforming Janus (0.61), [DALL-E 3](https://www.edgepedia.org/dall-e-3) (0.67), and Stable Diffusion 3 Medium (0.74).<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>
- **Prompt following (DPG-Bench):** 84.19, which the paper states surpasses all other methods in its comparison, including SD3-Medium (84.08) and DALL-E 3 (83.50).<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>
- **Understanding (MMBench):** 79.2 for the 7B model, versus Janus (69.4), TokenFlow (68.9), and MetaMorph (75.2).<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>
- **Detailed understanding scores (7B):** POPE 87.4, MME-P 1567.1, SEED 72.1, GQA 62.0, MMMU 41.0, MM-Vet 50.0; the 1B model scores [MMBench](https://www.edgepedia.org/mmbench) 75.5, MMMU 36.3, and GenEval 0.73.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>
- **GenEval sub-scores (7B):** Single Object 0.99, Two Objects 0.89, Counting 0.59, Colors 0.90, Position 0.79, Color Attributes 0.66; counting and color attributes are the weak spots.

In practice, the same review notes that DALL-E 3 still produces stronger photorealism, especially for human figures, while Janus-Pro is competitive on prompt adherence; the low native 384 × 384 resolution means dedicated diffusion models beat it on raw fidelity at 1024 × 1024.<sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup>

## How it compares with Janus, JanusFlow and rivals

Against its sibling Janus-1.3B, Janus-Pro raised GenEval from 0.61 to 0.80 and MMBench from 69.4 to 79.2, gains DeepSeek attributes to the enlarged training data and revised stage ratios.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> The 7B model also reportedly outperforms the larger 13B TokenFlow-XL on all understanding benchmarks except GQA.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup> The original Janus had already outperformed Show-o (53%), SDXL (55%), and [DALL-E 2](https://www.edgepedia.org/dall-e-2) (52%) on GenEval at 61%, and outperformed the 7B LLaVA-v1.5 and Qwen-VL-Chat on understanding benchmarks despite its smaller size.<sup>[4](https://arxiv.org/html/2410.13848)</sup>

Janus-Pro's native generation path is fully autoregressive rather than diffusion-based, placing it on one side of an ongoing design debate: unified autoregressive models (Show-o, [Chameleon](https://www.edgepedia.org/chameleon), Emu3, Janus-Pro) versus diffusion models (SDXL, SD3, DALL-E 3). The practical trade-off visible in the sources is that autoregressive token generation gives a single model both capabilities, while diffusion systems retain an edge in raw image fidelity at higher resolutions.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup><sup> • </sup><sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup>

## Reception, use and cost

Janus-Pro ships as open weights with no first-party API on api.deepseek.com; users run the weights themselves.<sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup> The 7B model loads in bf16 on a single 24 GB consumer GPU using [Hugging Face Transformers](https://www.edgepedia.org/hugging-face-transformers) and the official Janus loader, and the 1B variant runs in a browser via WebGPU, so local use requires a consumer gaming-class GPU rather than data-center hardware.<sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup> A review of the model's roles describes it as serving both as a unified autoregressive model for understanding and image generation and as a compact multimodal planner that can structure prompts for higher-resolution diffusion systems.

## Limitations and criticisms

The Janus-Pro paper itself acknowledges several limitations. Input resolution is limited to 384 × 384, which affects performance in fine-grained tasks such as OCR. The VQ tokenizer introduces reconstruction artifacts that leave small regions like faces under-detailed. The paper states it does not go deeply into safety mechanisms, and that curated aesthetic data and synthetic prompts can bias the model toward certain styles and subjects.<sup>[1](https://arxiv.org/pdf/2501.17811)</sup>

Third-party commentary identifies generating humans as a significant weakness: faces, hands, and full-body proportions are common failure modes, and dedicated diffusion models beat it on raw fidelity at 1024 × 1024.<sup>[5](https://deepseekai.guide/models/deepseek-janus/)</sup>

## References

1. [Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling (arXiv)](https://arxiv.org/pdf/2501.17811)
2. [deepseek-ai/Janus (official repository)](https://github.com/deepseek-ai/Janus)
3. [deepseek-ai/Janus-Pro-1B model card](https://huggingface.co/deepseek-ai/Janus-Pro-1B)
4. [Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation (arXiv)](https://arxiv.org/html/2410.13848)
5. [DeepSeek Janus: Multimodal Image Model Explained](https://deepseekai.guide/models/deepseek-janus/)

---
*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Modern AI: foundation models, generative AI, and the AI industry › Model families and named models › Multimodal, vision, and world models*

*Initially written Sep 17, 2026 · Edited: Oct 11, 2026; Oct 15, 2026*

*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*

License: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license