Zen Models - 95 open foundation models from 0.6B to 1T+

Zen Model Catalog

95 open foundation models across Zen3, Zen4, and Zen5

Chat, code, vision-language, web agentic, embeddings, rerankers, image generation, streaming ASR, and TTS. From edge-class Zen5 Nano 0.8B to the Zen5 Max frontier MoE. 8K - 1M context. OpenAI- and Anthropic-compatible API.

Browse on HuggingFace Get API Key Training Data

Zen 5 - Next-Generation Agentic

Native chain-of-thought, large-scale RL on 200K+ environments, OpenAI + Anthropic API.

Zen5 Nano 0.8B

Available

Parameters0.9B dense

ModalitiesMultimodal

Context32K

TargetEdge / on-device

Edge / on-device tier. Raspberry Pi, phone, browser WASM class.

SafeTensorsGGUFMLX

HuggingFace

Zen5 Nano 2B

Available

Parameters2B dense

ModalitiesMultimodal

Context32K

Target8 GB RAM laptop

Low-end multimodal dense at 2B parameters. Low iGPU / 8 GB RAM laptop class.

SafeTensorsGGUFMLX

HuggingFace

Zen5 Nano 4B

Available

Parameters5B dense

ModalitiesMultimodal

Context32K

Target16 GB / mobile NPU

Mid multimodal dense at 5B. 16 GB RAM laptop / mobile NPU class.

SafeTensorsGGUFMLX

HuggingFace

Zen5 Nano 9B

Available

Parameters10B dense

ModalitiesMultimodal

Context32K

Target24 GB+ / consumer GPU

Upper-nano multimodal dense. 24 GB+ unified RAM / consumer GPU class.

SafeTensorsGGUFMLX

HuggingFace

Zen5 Flash

Available

Parameters4B dense

ArchitectureZen dense

Context32K

TTFTSub-100ms

Smallest and cheapest text-only Zen5 chat tier. For high-volume routing and simple agent loops.

SafeTensors

HuggingFace

Zen5 Mini

Available

Parameters230B (10B active)

ArchitectureZen agentic MoE

Context192K

SWE-Bench80.2%

Frontier agentic at the lowest $/token in the family. 230B MoE / 10B active, trained on 200K+ real environments via large-scale RL.

SafeTensors

HuggingFace

Zen5

Available

Parameters35B (3B active)

ArchitectureZen frontier MoE

Context256K

LicenseApache-2.0

Canonical Zen5 default. 35B frontier MoE (3B active per token). The everyday Zen5 chat model. OpenAI + Anthropic API.

SafeTensorsGGUFMLX

HuggingFace

Zen5 Coder

Available

Parameters80B sparse MoE

SpecializationRepo-scale code

Context256K

FocusAgentic / tool-use

Code-specialized Zen5 tier. Sparse MoE tuned for repo-scale code understanding, agentic refactoring, and tool-use coding loops.

SafeTensorsGGUF

HuggingFace

Zen5 Pro

Available

Parameters284B (37B active)

ArchitectureZen Flash MoE

Context1M

QuantIQ2_XXS-imatrix (81 GB)

Zen Flash IQ2_XXS-imatrix weights (81 GB GGUF). 284B total / 37B active. Fits a single 128 GB Apple Silicon / DGX Spark / H100 80 GB.

GGUF

HuggingFace

Zen5 Max

Cloud Only

ParametersFull Zen Pro (432 GB)

ArchitectureZen Pro MoE

Context1M+

Hardware512 GB+ unified / 8x H100

Top quality tier in the family. Requires Mac Studio M3 Ultra 512 GB or 8x H100/H200 class GPU pool.

SafeTensors

HuggingFace

Zen 5 Embedding

Three-SKU embedding lineup served on /v1/embeddings.

Zen5 Embedding 0.6B

Available

Parameters0.6B

Dimensions1024

Context32K

Endpoint/v1/embeddings

Lightweight embedding model for high-throughput RAG and search.

Zen5 Embedding 4B

Available

Parameters4B

Dimensions2560

Context32K

Endpoint/v1/embeddings

Balanced embedding model for production RAG.

Zen5 Embedding 8B

Available

Parameters8B

Dimensions4096

Context32K

Endpoint/v1/embeddings

High-quality embeddings for production RAG, semantic search, and classification.

Zen 4 - Production Chat

The everyday Zen production line: MoE flagships, thinking models, and long-context.

Zen4 Mini

Available

ArchitectureDense

Context128K

TierStarter

UseFree tier / edge

Ultra-fast lightweight model optimized for speed and cost efficiency. Ideal for free tier.

HuggingFace

Zen4

Available

Parameters744B (40B active)

ArchitectureMoE

Context202K

TierUltra Max

Flagship MoE model for complex reasoning and multi-domain tasks.

HuggingFace

Zen4 Pro

Available

Parameters80B (3B active)

ArchitectureMoE

Context131K

TierUltra

Efficient MoE model for demanding workloads with strong reasoning at production-grade cost.

HuggingFace

Zen4 Thinking

Available

Parameters80B (3B active)

ArchitectureMoE + CoT

Context131K

TierPro Max

Dedicated reasoning model with explicit chain-of-thought capabilities.

HuggingFace

Zen4 Ultra

Available

Parameters744B (40B active)

ArchitectureMoE + CoT

Context262K

TierUltra Max

Maximum reasoning capability with extended chain-of-thought on MoE architecture.

HuggingFace

Zen4.1

Available

ArchitectureDense

Context1M

TierUltra

FocusLong-document / agentic

High-performance 1M context model for long-document analysis, large codebase reasoning, and agentic workflows.

HuggingFace

Zen4 Max

Available

ArchitectureDense

Context1M

TierUltra Max

FocusFrontier intelligence

Most capable model for complex reasoning, analysis, and agentic tasks. 1M token context window.

HuggingFace

Zen 4 Coder

Code-specialized MoE and dense models tuned for generation, review, debugging, and agentic programming.

Zen4 Coder Flash

Available

Parameters30B (3B active)

ArchitectureMoE

Context262K

TierPro Max

Lightweight code model optimized for speed and inline completions.

HuggingFace

Zen4 Coder

Available

Parameters480B (35B active)

ArchitectureMoE

Context163K

TierUltra

Code-specialized MoE model for generation, review, debugging, and agentic programming.

HuggingFace

Zen4 Coder Pro

Available

Parameters480B

ArchitectureDense BF16

Context131K

TierUltra Max

Full-precision BF16 code model for maximum accuracy on complex codebases.

HuggingFace

Zen 3 - Multimodal & Specialty

Vision, audio, web agentic, safety, and edge.

Zen3 Omni

Available

Parameters~200B

ModalitiesText + Vision + Audio

Context202K

ArchitectureDense Multimodal

Hypermodal model supporting text, vision, audio, and structured output.

HuggingFace

Zen3 VL

Available

Parameters30B (3B active)

ModalitiesVision + Language

Context262K

Sizes2B / 8B / 32B / 235B-A22B

Vision-language model for image understanding and visual reasoning. Default 30B-A3B MoE plus 2B, 8B, 32B, and frontier 235B-A22B variants.

HuggingFace

Zen3 Web

Available

Sizes8B / 14B / 32B

ArchitectureZen Web dense

Context32K

FocusBrowser agentic

Web-agentic models for browser automation, scraping, and on-page reasoning. Three tiers from edge to top-end.

HuggingFace

Zen3 Nano

Available

Parameters8B dense

Context128K

TierStarter

UseEdge / free tier

Ultra-lightweight model for edge deployment and low-latency tasks. Available on free tier.

HuggingFace

Zen3 Guard

Available

Parameters4B dense

Context65K

Categories9 safety

Languages119

Content safety classifier for moderation and guardrails. 9 safety categories, 119 languages.

HuggingFace

Zen 3 Embedding & Reranker

Text and multimodal embeddings plus rerankers for retrieval pipelines.

Zen3 Embedding

Available

Sizessmall / medium / default

Parameters0.6B / 4B / N/A

Context8K - 40K

Endpoint/v1/embeddings

High-quality text embeddings for RAG, search, and classification. OpenAI-compatible endpoint available.

Zen3 Reranker

Available

Sizessmall / medium / default

Parameters0.6B / 4B / 8B

Context40K

Endpoint/v1/rerank

High-quality rerankers for improving retrieval accuracy in RAG pipelines.

Zen3 VL Embedding

Available

Sizes2B / 8B

ModalitiesText + Image

Context32K

Endpoint/v1/embeddings

Multimodal embeddings (text + image) for vision-aware retrieval and semantic search.

Zen3 VL Reranker

Available

Sizes2B / 8B

ModalitiesText + Image

Context32K

Endpoint/v1/rerank

Vision-language rerankers for multimodal RAG. Reranks (query, image+text) pairs.

Zen 3 Image Generation

Eight image-generation SKUs from fast diffusion to broadcast-quality.

Zen3 Image

Available

TypeText-to-image + edit

TierPro Max

Pricing$0.04 / image

Endpoint/v1/images/generations

Best general-purpose image generation.

Zen3 Image Max

Available

TypeText-to-image

TierUltra Max

Pricing$0.08 / image

QualityMaximum

Maximum quality image generation for professional creative work.

Zen3 Image Fast

Available

TypeText-to-image

TierPro

Pricing$0.00035 / step

LatencyUltra-fast

Fastest image model for real-time generation.

Zen3 Image SDXL / Dev / Playground / SSD / JP

Available

Variants5 specialty models

Resolutionup to 1024px

Pricingfrom $0.00013/step

Endpoint/v1/images/generations

Specialized image models: SDXL (1024px), Dev (experimentation), Playground (aesthetic), SSD (fastest diffusion), JP (Japanese-specialized).

Zen 3 Audio & Speech

Speech-to-text, text-to-speech, streaming ASR, voice cloning, and forced alignment.

Zen3 Audio (STT)

Available

Variantsaudio / audio-fast

Languages100+

Pricingfrom $0.0012 / min

Endpoint/v1/audio/transcriptions

High-quality and fast speech-to-text transcription. 100+ languages.

Zen3 ASR (Streaming)

Available

Variantsasr / asr-0.6B / asr-aligner / asr-v1

LatencySub-200ms - Sub-500ms

Pricingfrom $0.002 / min

Endpoint/v1/audio/transcriptions

Real-time streaming ASR for voice agents. Edge variant (0.6B) for on-device, aligner for word-level timestamps.

Zen3 TTS

Available

Variantstts / tts-hd / tts-fast / tts-0.6B

Voices40+ across 8 languages

Pricingfrom $2 / 1M chars

Endpoint/v1/audio/speech

High-quality text-to-speech with natural prosody. Four tiers from edge to broadcast-grade HD.

Zen3 TTS Voice Design & Custom Voice

Available

Variantsvoice-design / custom-voice

FeaturesPrompt-driven + few-shot clone

Pricing$8 - $10 / 1M chars

Endpoint/v1/audio/speech

Premium TTS with prompt-driven voice design and few-shot voice cloning from a short audio sample.

Full Zen Catalog Summary

Live catalog from the Zen API. Pricing fetched at runtime.

Generation	Family	SKUs	Endpoint(s)
Zen 5	Chat ladder	10 (nano 0.8B / 2B / 4B / 9B, flash, mini, default, coder, pro, max)	`/v1/chat/completions`
Zen 5	Embedding	3 (0.6B / 4B / 8B)	`/v1/embeddings`
Zen 4	Chat	7 (mini, default, pro, thinking, ultra, 4.1, max)	`/v1/chat/completions`
Zen 4	Coder	3 (flash, coder, pro)	`/v1/chat/completions`
Zen 3	Chat & VL	10+ (omni, nano, guard, vl x5, web x3)	`/v1/chat/completions`
Zen 3	Embedding & Reranker	11 (text + multimodal embeddings, rerankers)	`/v1/embeddings`, `/v1/rerank`
Zen 3	Image	8 (image, max, dev, fast, sdxl, playground, ssd, jp)	`/v1/images/generations`
Zen 3	Audio	6 STT/ASR + 6 TTS	`/v1/audio/transcriptions`, `/v1/audio/speech`

The Complete Catalog

All 95 open Zen models — every one linked to its weights on HuggingFace and its paper.

Chat & Reasoning

Zen 51M · text-generation

Weights Paper

Zen 5 Flash4.02B (dense) · 32K · text-generation

Weights Paper

Zen 5 MaxMoE (full-Pro base, IQ2_XXS quant) · 1M · text-generation

Weights Paper

Zen 5 Mini256K · text-generation

Weights Paper

Zen 5 Pro512K · text-generation

Weights Paper

Zen 5 Pro284B total / 37B active (MoE) · 1M · text-generation

Weights Paper

Zen Blog8.19B · 128K · text-generation

Weights Paper

Zen Eco0.75B · 32K · text-generation

Weights Paper

Zen Eco 4B Instruct4.02B · 32K · text-generation

Weights Paper

Zen Eco 4B Thinking4.02B · 32K · text-generation

Weights Paper

Zen Eco Instruct4B · 32K · text-generation

Weights Paper

Zen Multilingual8B · 128K · text-generation

Weights Paper

Zen Nano0.6B · 32K · text-generation

Weights Paper

Zen Nano 0.6B0.6B · 32K · text-generation

Weights Paper

Zen Pro8.19B · 128K · text-generation

Weights Paper

Zen Scribe2.35B · 32K · text-generation

Weights Paper

Zen3 Nano8.19B · 40K · text-generation

Weights Paper

Code

Zen 5 Coder256K · text-generation

Weights Paper

Zen 5 Coder79.7B (MoE) · text-generation

Weights Paper

Zen SQL8B · 32K · text-generation

Weights Paper

Vision-Language

Zen 535B total / 3B active (MoE) · 256K · image-text-to-text

Weights Paper

Zen 5 Nano 0.8B0.87B (dense) · image-text-to-text

Weights Paper

Zen 5 Nano 2B2.27B (dense) · image-text-to-text

Weights Paper

Zen Designer235B-A22B · 128K · visual-question-answering

Weights Paper

Zen Designer (GGUF)235B-A22B · 256K · text-generation

Weights Paper

Zen Designer 235B A22B Instruct236B · 131K · visual-question-answering

Weights Paper

Zen Designer 235B A22B Thinking236B · 131K · visual-question-answering

Weights Paper

Zen VL 30B Agent30B · 256K · image-text-to-text

Weights Paper

Zen VL 30B Instruct30B · 256K · image-text-to-text

Weights Paper

Zen VL 4B Agent4B · 32K · image-text-to-text

Weights Paper

Zen VL 4B Instruct4B · 32K · image-text-to-text

Weights Paper

Zen VL 8B Agent8B · 32K · image-text-to-text

Weights Paper

Zen VL 8B Instruct8B · 32K · image-text-to-text

Weights Paper

Zen3 VL8.77B (30B MoE) · 262K · text-generation

Weights Paper

Zen5 Nano 4B (GGUF)4.66B · image-text-to-text

Weights Paper

Zen5 Nano 9B (GGUF)9.65B · image-text-to-text

Weights Paper

Omni / Multimodal

Zen Omni30B (3B active MoE) · 32K · image-text-to-text

Weights Paper

Zen Omni 30B Instruct35.3B · 128K · text-generation

Weights Paper

Zen Omni 30B Thinking31.7B · 128K · image-text-to-text

Weights Paper

Zen3 Omni35.3B (1T MoE) · 202K · text-generation

Weights Paper

Embeddings

Zen Embedding7.57B · 8K · feature-extraction

Weights Paper

Zen Embedding 0.6B0.6B · 32K · feature-extraction

Weights Paper

Zen Embedding 0.6B (GGUF)0.6B · 8K · feature-extraction

Weights Paper

Zen Embedding 4B4.02B · 32K · feature-extraction

Weights Paper

Zen Embedding 8B7.57B · 32K · feature-extraction

Weights Paper

Zen Embedding 8B (GGUF)8B · 8K · feature-extraction

Weights Paper

Rerankers

Zen Reranker4.02B · 32K · text-classification

Weights Paper

Zen Reranker 0.6B0.6B · 32K · text-classification

Weights Paper

Zen Reranker 0.6B (GGUF)0.6B · 8K · text-classification

Weights Paper

Zen Reranker 4B4.02B · 32K · text-classification

Weights Paper

Zen Reranker 4B (GGUF)4B · 8K · text-classification

Weights Paper

Zen Reranker 8B8.19B · 32K · text-classification

Weights Paper

Zen Reranker 8B (GGUF)8B · 8K · text-classification

Weights Paper

Safety

Zen Guard3.09B · text-classification

Weights Paper

Zen Guard Gen7.62B · 32K · text-classification

Weights Paper

Zen Guard Gen 8B7.62B · text-generation

Weights Paper

Zen Guard Stream3B · text-classification

Weights Paper

Zen Guard Stream 4B3.09B · text-generation

Weights Paper

Zen3 Guard5.78B · text-classification

Weights Paper

Image

Zen 3 Image JP12B · text-to-image

Weights Paper

Zen 3 Image SSD12B · text-to-image

Weights Paper

Zen Image Edit7B · image-to-image

Weights Paper

Zen3 Image12B · 1024px · text-to-image

Weights Paper

Zen3 Image Dev12B · 1024px · text-to-image

Weights Paper

Zen3 Image Fast12B · 1024px · text-to-image

Weights Paper

Zen3 Image Max12B · 2048px · text-to-image

Weights Paper

Zen3 Image Playground12B · 1024px · text-to-image

Weights Paper

Zen3 Image SDXL12B · 1024px · text-to-image

Weights Paper

Video

Zen Director5B · text-to-video

Weights Paper

Zen Video13B · text-to-video

Weights Paper

Zen Video I2V13B · image-to-video

Weights Paper

Audio & Speech

Zen Dub Live1.93B · 30s audio · audio-to-audio

Weights Paper

Zen Foley1B · 10s audio · text-to-audio

Weights Paper

Zen Musician6.22B · 30s audio · text-to-audio

Weights Paper

Zen3 ASR2.35B · automatic-speech-recognition

Weights Paper

Zen3 ASR 0.6B0.94B · automatic-speech-recognition

Weights Paper

Zen3 ASR Forced Aligner0.92B · automatic-speech-recognition

Weights Paper

Zen3 TTS1.93B · text-to-speech

Weights Paper

Zen3 TTS 0.6B0.91B · text-to-speech

Weights Paper

Zen3 TTS Custom Voice1.92B · text-to-speech

Weights Paper

Zen3 TTS Voice Design1.92B · text-to-speech

Weights Paper

3D

Zen 3Dimage-to-3d

Weights Paper

Agents

Zen 5 MiniMoE (~10B active) · text-generation

Weights Paper

Zen Agent 4B4.02B · 32K · text-generation

Weights Paper

Zen Eco 4B Agent (GGUF)4.02B · 32K · text-generation

Weights Paper

Zen Eco 4B Agent (MLX)4.02B · 32K · text-generation

Weights Paper

World Models

Zen Voyager32.8B · text-to-video

Weights Paper

Zen World13B · text-to-video

Weights Paper

Specialty

Zen Family8.19B · text-generation

Weights Paper

Zen Finance8B · 32K · text-generation

Weights Paper

Zen Legal8B · 131K · text-generation

Weights Paper

Zen Medical8B · 32K · text-generation

Weights Paper

Zen Trainingtext-generation

Weights Paper

Zen Translate8B · 32K · text-generation

Weights Paper

Zen Translator7B · translation

Weights Paper