Zen AI Model Family

Hypermodal AI

45+ models across 13 families. Open-weight AI covering text, vision, image, audio, code, embeddings, and reranking — from edge to frontier.

45+
AI Models
13
Model Families
7
Modalities
Apache 2.0
License

Zen 5

Flagship generation — sparse mixture-of-experts for text, code, and math.

zen5

35B (3B active)

Flagship model. Sparse mixture-of-experts, 35B total / 3B active per token. Open weights are a GGUF repackage of Qwen3.6-35B-A3B (Apache-2.0).

TextCodeMath
1M contextMoE

zen5-mini

TBA

Fast, low-cost tier. Agentic MoE with ~10B active parameters per token. Open weights are a GGUF repackage of MiniMax-M2.5 (modified-MIT).

TextCode
1M contextMoE

zen5-coder

80B (3B active)

Code-specialized MoE, 80B total / 3B active per token. Open weights are a GGUF repackage of Qwen3-Next-80B-A3B-Instruct (Apache-2.0).

TextCode
1M contextMoE

zen5-pro

TBA

Higher-capability tier for demanding production workloads. The served SKU is full-precision in the cloud; the linked repo is a low-bit (IQ2XXS) GGUF re-host of DeepSeek-V4-Flash for local runs.

TextCodeMath
1M contextMoE

zen5-max

TBA

Maximum-capability tier for the hardest reasoning and agentic tasks. The served SKU is full-precision in the cloud; the linked repo is a low-bit (IQ2XXS) GGUF re-host of DeepSeek-V4-Pro for local runs.

TextCodeMath
1M contextMoE

Zen 3

Multimodal, vision-language, and safety models.

zen3-omni

30B (3B active)

Multimodal model (text + vision), 30B total / 3B active MoE. Open weights repackage Qwen3-Omni-30B-A3B-Instruct (Apache-2.0).

TextVision
131K contextMoE Vision-Language

zen3-vl

8B

Vision-language model for image understanding, 8B dense. Open weights repackage Qwen3-VL-8B-Instruct (Apache-2.0).

TextVision
128K contextDense

zen3-guard

5B

Safety classifier for content moderation and guardrails, ~5B dense. Open weights repackage ibm-granite/granite-guardian-3.2-5b (Apache-2.0).

TextSafety
128K contextDense

Code

Code generation, review, and agentic programming.

zen5-coder

80B (3B active)

Code-specialized MoE, 80B total / 3B active per token. Open weights are a GGUF repackage of Qwen3-Next-80B-A3B-Instruct (Apache-2.0).

TextCode
1M contextMoE

Vision

Vision-language and multimodal understanding.

zen3-omni

30B (3B active)

Multimodal model (text + vision), 30B total / 3B active MoE. Open weights repackage Qwen3-Omni-30B-A3B-Instruct (Apache-2.0).

TextVision
131K contextMoE Vision-Language

zen3-vl

8B

Vision-language model for image understanding, 8B dense. Open weights repackage Qwen3-VL-8B-Instruct (Apache-2.0).

TextVision
128K contextDense

Embedding & Retrieval

Text embeddings and reranking — served endpoints and open weights.

zen5-embedding

TBA
Research Preview

Text-embedding endpoint in the Zen5 line for RAG, search, and classification. Served via API.

EmbeddingsRAG
33K contextEmbedding

zen3-embedding

TBA
Research Preview

Text-embedding endpoint in the Zen3 line for RAG, search, and classification. Served via API; open weights available under the zen-embedding-* repos.

EmbeddingsRAG
33K contextEmbedding

zen-rerank

TBA
Research Preview

Reranking endpoint that re-scores candidate passages for retrieval pipelines. Served via API, billed per unit; open weights available under the zen-reranker-* repos.

EmbeddingsReranking
RerankerReranker

zen-embedding-8b

8B

8B text-embedding model for dense retrieval. Open-weight repackage of Qwen/Qwen3-Embedding-8B (Apache-2.0).

Embeddings
41K contextEmbedding

zen-embedding-4b

4B

4B text-embedding model for dense retrieval. Open-weight repackage of Qwen/Qwen3-Embedding-4B (Apache-2.0).

Embeddings
41K contextEmbedding

zen-embedding-0.6b

0.6B

0.6B text-embedding model for high-throughput retrieval. Open-weight repackage of Qwen/Qwen3-Embedding-0.6B (Apache-2.0).

Embeddings
33K contextEmbedding

zen-reranker-8b

8B

8B cross-encoder reranker for search relevance. Open-weight repackage of Qwen/Qwen3-Reranker-8B (Apache-2.0).

EmbeddingsReranking
41K contextReranker

zen-reranker-4b

4B

4B cross-encoder reranker for search relevance. Open-weight repackage of Qwen/Qwen3-Reranker-4B (Apache-2.0).

EmbeddingsReranking
41K contextReranker

zen-reranker-0.6b

0.6B

0.6B cross-encoder reranker for high-throughput reranking. Open-weight repackage of Qwen/Qwen3-Reranker-0.6B (Apache-2.0).

EmbeddingsReranking
41K contextReranker

Image Generation

Text-to-image and image editing diffusion — served endpoint and open weights.

zen-image

TBA
Research Preview

Text-to-image diffusion endpoint. Served via API, billed per image; open weights available under zen3-image.

Vision
DiffusionDiffusion

zen3-image

TBA

Text-to-image diffusion model. Open-weight repackage of Qwen/Qwen-Image (Apache-2.0).

Vision
DiffusionDiffusion

zen-image-edit

TBA

Instruction-based image editing diffusion model. Open-weight repackage of Qwen/Qwen-Image-Edit-2509 (Apache-2.0).

Vision
DiffusionDiffusion

Audio & Speech

Text-to-speech, text-to-music, sound effects, speech recognition, and dubbing.

zen-voice

TBA
Research Preview

Text-to-speech endpoint. Served via API, billed per unit; open weights available under zen3-tts.

Audio
TTSTTS

zen-music

TBA
Research Preview

Text-to-music endpoint. Served via API, billed per unit; open weights available under zen-musician.

Audio
DenseDense

zen-foley

TBA
Research Preview

Text-to-sound-effects endpoint. Served via API, billed per unit.

Audio
DiffusionDiffusion

zen3-tts

1.7B

Text-to-speech model, ~1.7B, 12 Hz codec, multilingual synthesis. Fine-tuned from Qwen/Qwen3-TTS-12Hz-1.7B-Base (Apache-2.0).

AudioMultilingual
TTSTTS

zen3-asr

1.7B

Speech-recognition model, ~1.7B, 24-layer audio encoder feeding a Qwen3 decoder; 30 languages. Fine-tuned from Qwen/Qwen3-ASR-1.7B (Apache-2.0).

TextAudioStreaming
ASRASR

zen-musician

7B

Text-to-music model, 7B autoregressive. Repackage of m-a-p/YuE-s1-7B-anneal-en-cot (Apache-2.0).

Audio
DenseDense

zen-dub-live

TBA

Live dubbing / text-to-speech model. Fine-tuned from the Qwen3-TTS family (Apache-2.0).

Audio
TTSTTS

Video Generation

Text-to-video and image-to-video diffusion.

zen-video

TBA
Research Preview

Text-to-video diffusion endpoint. Served via API, billed per unit; related open weights: zen-video-i2v (Wan2.2).

DiffusionDiffusion

zen-world

14B

Text-to-video diffusion model, 14B. Open-weight repackage of Wan-AI/Wan2.1-T2V-14B (Apache-2.0).

DiffusionDiffusion

zen-director

5B

Text + image to video diffusion model, 5B. Open-weight repackage of Wan-AI/Wan2.2-TI2V-5B (Apache-2.0).

Vision
DiffusionDiffusion

Safety

Content moderation and safety classification.

zen3-guard

5B

Safety classifier for content moderation and guardrails, ~5B dense. Open weights repackage ibm-granite/granite-guardian-3.2-5b (Apache-2.0).

TextSafety
128K contextDense

Foundation (Open Weights)

General-purpose open-weight text models.

zen3-nano

8B

Compact 8B dense text model. Open-weight repackage of Qwen/Qwen3-8B (Apache-2.0).

TextCode
41K contextDense

zen-eco

4B

Efficient 4B dense text model. Open-weight repackage of Qwen/Qwen3-4B (Apache-2.0).

TextCode
262K contextDense

zen-nano

0.6B

0.6B dense text model for edge and mobile. Open-weight repackage of Qwen/Qwen3-0.6B (Apache-2.0).

TextCode
41K contextDense

Agents

Open-weight models tuned for tool use and multi-step planning.

zen-agent-4b

4B

Agent / tool-use model, 4B dense. Open-weight repackage of Qwen/Qwen3-4B (Apache-2.0); shares the Qwen3-4B lineage with zen-eco, tuned for multi-step tool use.

TextCodeAgents
262K contextDense

zen-router

0.6B

Tiny routing model: reads a prompt and picks the best model to answer it. One forward pass emits a task class, a route distribution over the model catalog, and a feature embedding.

TextAgents
32K contextDense

Domain Verticals

Domain-specialized open-weight text models.

zen-medical

32B

Medical-domain text model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

Text
41K contextDense

zen-legal

32B

Legal-domain text model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

Text
41K contextDense

zen-finance

32B

Finance-domain text model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

Text
41K contextDense

zen-sql

32B

Text-to-SQL model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

TextCode
41K contextDense

zen-translate

32B

Translation model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

Text
41K contextDense

zen-multilingual

32B

Multilingual text model, 32B dense. Fine-tune of zen-pro (Qwen3-32B lineage, Apache-2.0).

TextMultilingual
41K contextDense

Creative Studio

Open-weight video, image editing, 3D, dubbing, and design models.

zen-world

14B

Text-to-video diffusion model, 14B. Open-weight repackage of Wan-AI/Wan2.1-T2V-14B (Apache-2.0).

DiffusionDiffusion

zen-director

5B

Text + image to video diffusion model, 5B. Open-weight repackage of Wan-AI/Wan2.2-TI2V-5B (Apache-2.0).

Vision
DiffusionDiffusion

zen-image-edit

TBA

Instruction-based image editing diffusion model. Open-weight repackage of Qwen/Qwen-Image-Edit-2509 (Apache-2.0).

Vision
DiffusionDiffusion

zen-3d

TBA

Image-to-3D asset generation model. Open-weight repackage of JeffreyXiang/TRELLIS-image-large (MIT).

Vision
DiffusionDiffusion

zen-dub-live

TBA

Live dubbing / text-to-speech model. Fine-tuned from the Qwen3-TTS family (Apache-2.0).

Audio
TTSTTS

zen-designer-235b-a22b-instruct

235B (22B active)

Vision-language design model, 235B total / 22B active MoE. Open-weight repackage of Qwen/Qwen3-VL-235B-A22B-Instruct (Apache-2.0).

TextVision
131K contextMoE Vision-Language

zen-designer-235b-a22b-thinking

235B (22B active)

Reasoning vision-language design model, 235B total / 22B active MoE. Open-weight repackage of Qwen/Qwen3-VL-235B-A22B-Thinking (Apache-2.0).

TextVisionReasoning
131K contextMoE Vision-Language

zen-designer-gguf

235B (22B active)

GGUF build of Zen Designer for local runs, 235B total / 22B active MoE. Repackage of Qwen/Qwen3-VL-235B-A22B-Instruct (Apache-2.0).

TextVision
131K contextMoE Vision-Language

Capabilities Matrix

Each model specializes in different modalities and tasks

ModelTextImageVideoAudio3DCodeAgents
zen5✓————✓✓
zen4✓————✓✓
zen4-max✓————✓✓
zen4-ultra✓————✓✓
zen4-coder✓————✓✓
zen3-omni✓✓—✓———
zen3-vl✓✓—————
zen3-nano✓————✓—
zen3-guard✓——————
zen3-image✓✓—————
zen3-audio✓——✓———
zen3-tts✓——✓———
zen3-embedding✓——————
zen3-reranker✓——————

Infrastructure

Production-ready tools for training and deploying Zen models

Hanzo Engine

High-performance cloud inference — 60+ architectures, CUDA/Metal, OpenAI-compatible API

Hanzo Edge

On-device AI inference — run models locally on any device, browser, or embedded system

Zen Gym

Unified training platform for all Zen models with LoRA, QLoRA, GRPO, and more

Hanzo MCP

Model Context Protocol for AI context management and tool use

Quick Start

Get started with any Zen model in seconds

# Install and run any model
pip install transformers torch

# Use directly
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("zenlm/zen-eco-4b-instruct")
tokenizer = AutoTokenizer.from_pretrained("zenlm/zen-eco-4b-instruct")

# Or use via Hanzo Cloud API
from hanzoai import Hanzo
client = Hanzo(api_key="hk-your-api-key")

response = client.chat.completions.create(
    model="zen-eco-4b-instruct",
    messages=[{"role": "user", "content": "Hello!"}]
)

Build with Zen AI

Open-weight models, Apache 2.0 licensed. Free to use for research and commercial applications.