# The Zen family — Hanzo AI

> Open-weight foundation models for agentic coding. Download weights or run Zen 6 and Zen 6 Flash directly on Hanzo Cloud.

Open models

# Zen 6

Open models. Yours to run.

The open-weight family behind Hanzo, from a laptop to the datacenter. Download the weights, run them privately, or call them on Hanzo Cloud.

[Get API key](https://platform.hanzo.ai/keys)[Download the weights](#download)

Available now

## Zen 6

27.3B dense. 262K context, 1M with YaRN. Text, images and video. Zen 6 Flash packs it into 5.95 GB for a laptop.[Zen 6](https://hanzo.ai/models/hanzo/zen6)[Zen 6 Flash](https://hanzo.ai/models/hanzo/zen6-flash)[Run it](#run)

Research preview

## Zen 7

The next generation, open to researchers by request. No public weights and no callable id yet.[Request access](https://hanzo.ai/research-access?track=zen)[The lineage](#lineage)

Earlier Zen models[zen-embedding](https://hanzo.ai/models/hanzo/zen-embedding)[zen-free](https://hanzo.ai/models/hanzo/zen-free)[zen-guard](https://hanzo.ai/models/hanzo/zen-guard)[zen-scribe](https://hanzo.ai/models/hanzo/zen-scribe)[zen-vl](https://hanzo.ai/models/hanzo/zen-vl)[zen-voice-mini](https://hanzo.ai/models/hanzo/zen-voice-mini)[zen5](https://hanzo.ai/models/hanzo/zen5)[zen5-coder](https://hanzo.ai/models/hanzo/zen5-coder)[zen5-evo](https://hanzo.ai/models/hanzo/zen5-evo)[zen5-flash](https://hanzo.ai/models/hanzo/zen5-flash)[zen5-mini](https://hanzo.ai/models/hanzo/zen5-mini)[zen5-pro](https://hanzo.ai/models/hanzo/zen5-pro)[zen5-spark](https://hanzo.ai/models/hanzo/zen5-spark)

[The full Zen catalogue](https://hanzo.ai/zen/models)

Available now

## Zen 6.Open now.

A 27.3 billion parameter dense model. Hybrid attention, a long context, images and video, and a speculative drafter bundled with the weights. The interesting part is not that it is the biggest. It is that you can run it.

### Agentic coding

Long repositories, tool use, long context, and fast speculative decoding.

### Multimodal work

Zen 6 reads text, images, and video in one context.

27.3BParameters

262KNative context

1MWith YaRN

On one DGX Spark, the weight card records 62.4 tokens per second on code alone and 141.2 with the drafter. That figure is published with the weights. It is not a Hanzo harness.[Zen 6 Flash](#flash)[zenlm/zen6](https://huggingface.co/zenlm/zen6)

Zen 6 Flash

## 27B.Laptop-sized.

The same generation, packed ternary. 26,895,998,464 language-model weights in a 5.95 GB file, 1.77 bits per weight, 9.0 times smaller than the 53.79 GB FP16 representation. Text and images. Not video.

FP16 representation53.79 GB

Ternary1.77 bits / weight

5.95 GBDense packed

5.95 GBDense packed

1.77Bits per weight

262KNative context

Apple Silicon or one GPU, through llama.cpp and Metal. The card records 47.2 tokens per second alone and 92.4 with the drafter on M4 and M5 Max, in 7.84 GB with the projector and a 32K cache. Published with the weights. Not a Hanzo harness.[zenlm/zen6-flash](https://huggingface.co/zenlm/zen6-flash)

Measured by Hanzo

## A frontier model.On your desk.

Hanzo ran Zen 6 on three machines it owns. These are the readings, in tokens a second.

### NVIDIA Grace-Blackwell

121 GB unified · NVFP4 weights

~2,400tok/s prefill

~43tok/s decode

50.6 tok/s decoding code · 112 tok/s across 8 requests

### AMD Strix Halo

128 GB unified · 4-bit build

1,150–1,400tok/s prefill

~34tok/s decode

262K context, in two slots

### Apple silicon

unified memory · MLX build

500–690tok/s prefill

64–81tok/s decode

Hanzo's runs, 23 September 2026. Decode is one request unless a line says otherwise, and each machine runs its own build of Zen 6. Where the Hanzo engine's prefill on CUDA beats llama.cpp is on the inference benchmark.[Engine prefill on CUDA](https://hanzo.ai/benchmarks/inference)

Open weights

## APIs can disappear.Weights don't.

Download them, read them, fine-tune them, and run them where you want. Zen 6 is Apache-2.0, published by Zoo Labs Foundation, a 501(c)(3) non-profit.

Zen weightsYour laptopYour GPUHanzo Cloud

[The family on Hugging Face](https://huggingface.co/zenlm)[Source](https://github.com/zenlm)

2023 to now

## Seven generations. One line.

Foundation, then context, then a family, then scale, then the models that actually shipped. Each generation inherits the last, and the current one is the practical one.

01

02

03

04

05

5.8

06

07

01

### Zen 1

Q4 2023

HistoricalFoundationDense multilingual experiments at 7B, 14B and 72B, to prove an open foundation could scale. These are not models the gateway serves.One dense block.

02

### Zen 2

Q2 2024

HistoricalGo longer128K context, the first production deployments, and code and math specialists. From research model to production model. The context got longer. The architecture did not get more decorative.The same block, wider.

03

### Zen 3

2024

Open weightsMore than languageThe line branched: sparse experts, vision, audio, image, retrieval and safety. This is where Zen becomes a family. None of these ids are in the live catalog. Open-weight specialist artifacts are still published.One line, six branches.

LanguageVisionAudioImageRetrievalSafety

[zenlm/zen3-omni](https://huggingface.co/zenlm/zen3-omni)[zenlm/zen3-tts](https://huggingface.co/zenlm/zen3-tts)[zenlm/zen3-guard](https://huggingface.co/zenlm/zen3-guard)

04

### Zen 4

2025

RetiredScale upFrontier-scale experts and a longer context. Retired. Not a current production model, and not a public download.Large, and sparse.

05

### Zen 5

ShippedAgentic generationCode, long context, and smaller and host-specific variants. Older pages described a much larger routed checkpoint. It was not published. These are the models that were.Paths for tools and hosts.

zen5 zen5-coder zen5-evo zen5-flash zen5-mini zen5-pro zen5-spark

5.8

### Zen 5.8

Served onlyServing bridgeA served-only bridge between Zen 5 and Zen 6, with no public weights. The gateway does not list it now.A thin serving step.

06

### Zen 6

CurrentDense againZen 6 is a practical dense model: hybrid attention, a long context, images and video, speculative decoding, and Zen 6 Flash, a ternary build that fits a laptop. Available now.Smaller, and solid.

zen6 zen6-flash[The current generation](#zen6)

07

### Zen 7

Research previewResearch previewZen 7 has no public weights and no callable id. Its architecture, size, modalities and date are not stated. Research preview — not currently available.The release path, unresolved past the preview.

What comes next is still research.

Train

Attest

EvaluateNot published

ReleaseNot published

The path to a release is training, then attestation, then evaluation, then publication. Nothing past the preview is published.[Request research access](https://hanzo.ai/research-access?track=zen)[The family on Hugging Face](https://huggingface.co/zenlm)

Two artifacts

## Open weights, and a hosted id.

The same name can be two things. A hosted id is not documented as the same bytes as a public repository with a similar name.

### Open weights

Download. Self-host. Modify.

### Hosted

An API call. Managed. Metered.

Call it

## One call.Any SDK.

The gateway speaks the OpenAI API, so the SDK you already use works. Point it at api.hanzo.ai and ask for "zen6". The weights are the download.

python

```
from openai import OpenAI client = OpenAI( base_url="https://api.hanzo.ai/v1", api_key="hk-...", # your Hanzo API key ) resp = client.chat.completions.create( model="zen6", messages=[{"role": "user", "content": "Hello"}], ) print(resp.choices[0].message.content)
```

bash

```
curl https://api.hanzo.ai/v1/chat/completions \ -H "Authorization: Bearer $HANZO_API_KEY" \ -H "Content-Type: application/json" \ -d '{"model": "zen6", "messages": [{"role": "user", "content": "Hello"}]}'
```

[Try in Hanzo](https://hanzo.ai/signup)Copy API call

[Download weights](#download)

Run it yourself

## Three ways.

Each command runs against the weights as published.

### Download

hf download zenlm/zen6Copy

hf download zenlm/zen6-flashCopy

### Local

Zen 6 Flash. llama.cpp and Metal. Apple Silicon or one GPU.bash

```
hf download zenlm/zen6-flash --local-dir zen6-flash llama-server \ -m zen6-flash/*-PQ2_0.gguf \ --mmproj zen6-flash/*-mmproj-Q8_0.gguf \ -md zen6-flash/*-DFlash2-Q8_0.gguf \ --spec-draft-n-max 3 \ -c 32768 \ --port 8080
```

### Production

Zen 6. SGLang, with the bundled drafter.bash

```
hf download zenlm/zen6 --local-dir zen6 python3 -m sglang.launch_server \ --model-path ./zen6 \ --speculative-draft-model-path ./zen6/dflash2 \ --speculative-num-steps 3 \ --speculative-algorithm DFLASH \ --kv-cache-dtype fp8_e5m2 \ --context-length 1048576 \ --port 30000 \ --host 0.0.0.0
```

Specifications

## Zen 6 beside Zen 6 Flash.

Card facts only. A cell with no published value is left blank.

Zen 6Zen 6 Flash

ParametersZen 627.3 billion, dense, plus a vision encoder

Zen 6 Flash26,895,998,464 language-model weights

Weight formatZen 6NVFP4 on the MLP and LM head, FP8 on attention

Zen 6 FlashTernary, 1.77 bits per weight, 5.95 GB packed

Native contextZen 6262,144

Zen 6 Flash262,144

Extended contextZen 61,048,576 with YaRN

Zen 6 Flash—

TextZen 6Yes

Zen 6 FlashYes

ImagesZen 6Yes

Zen 6 FlashYes

VideoZen 6Yes

Zen 6 Flash—

Runs onZen 6A GPU, with SGLang. Measured on one DGX Spark.

Zen 6 FlashA laptop or one GPU, with llama.cpp and Metal

The API catalog lists 1M for zen6 and 1M for zen6-flash. The weight cards state 262,144 tokens native, and YaRN to 1,048,576 for Zen 6 only.Layers, heads, and the files

Zen 6 has 64 layers, 48 of them linear attention and 16 full attention, one full layer every fourth. Hidden size 5,120. 24 query heads and 4 key and value heads, head size 256. Vocabulary 248,320. The speculative drafter ships in the weights.Zen 6 Flash also publishes a 7.21 GB file at 2.14 bits per weight, which is the one the kernels read, a 629 MB vision projector, and a 2.06 GB drafter. With the projector and a 32K cache it stays under 8 GB.

Scored

## Scored whereit counts.

The publisher scores the open weights. A Hanzo score of the family is a research run, and none is filed yet, so this page shows none. The two are never mixed.

98.2%Kept by Zen 6 Flash's ternary build

The weights report 86.33 at full precision and 84.78 for the ternary build across 14 reasoning suites: the publisher's numbers, not a Hanzo harness. Zen 6 is not on the board yet, so this page does not borrow one.[Every benchmark](https://hanzo.ai/benchmarks)

## Your model shouldsurvive your vendor.

Weights you can hold, runtimes you can choose, infrastructure you control.

DownloadInspectFine-tuneDeployFork

Zen · Foundation Models

## Open-weight intelligence. Full parameter specifications, endpoints, and runtimes.

Dense 27.3B foundation architecture with bundled speculative drafter, ternary quantization for edge deployment, 1M context with YaRN, and OpenAI-compatible streaming SSE.

Model Card Specifications

zen6

Wire EndpointPOST https://api.hanzo.ai/v1/chat/completions

Parameters

27.3 Billion (dense) + vision

Quantization5.95 GB packed for edge deployment

NVFP4 + FP8 · Ternary (Flash)

Context Reach1M tokens supported on Hanzo Cloud

262,144 native · 1,048,576 with YaRN

ThroughputWith bundled speculative drafter

141.2 tokens/sec (DGX Spark)

Streaming ProtocolIncludes reasoning_content tokens

Server-Sent Events (SSE)

Local Runtimes

llama.cpp · SGLang · vLLM · Metal

Weights & Licensinghuggingface.co/zenlm

Apache-2.0 on Hugging Face

[Get API key](https://platform.hanzo.ai/keys)[🤗 zenlm/zen6 on HF Hub](https://huggingface.co/zenlm/zen6)[Documentation](https://docs.hanzo.ai/docs/models/zen)

Multi-Language SDKs & OSS Examples

[🤗 Hugging Face](https://huggingface.co/zenlm/zen6)[GitHub Repo](https://github.com/hanzoai/zen)

Runnable examples across all tier-one runtimes. Zero-install authentication with hanzo auth login or environment variable HANZO_API_KEY.

[Pythonexamples/cloud](https://github.com/hanzoai/zen/tree/main/examples/cloud/client_cloud.py)

[Goexamples/cloud](https://github.com/hanzoai/zen/tree/main/examples/cloud/client_cloud.go)

[Rustexamples/cloud](https://github.com/hanzoai/zen/tree/main/examples/cloud/client_cloud.rs)

[TypeScriptexamples](https://github.com/hanzoai/zen/tree/main/examples)

[cURLcurl](https://github.com/hanzoai/zen/tree/main/examples/cloud)

[Modular Rust Recipes CrateModular production blueprints, enterprise crates, and automated verification suites.](https://github.com/hanzoai/zen/tree/main/recipes)

python

```
import os from openai import OpenAI client = OpenAI( base_url="https://api.hanzo.ai/v1", api_key=os.environ["HANZO_API_KEY"], ) response = client.chat.completions.create( model="zen6", messages=[{"role": "user", "content": "Write a binary search in Rust."}], stream=True, ) for chunk in response: print(chunk.choices[0].delta.content or "", end="")
```

Read

## Models, benchmarks, research.

The weights you can run, where they are scored, and the writing.

### Models

[Zen 6](#zen6)

[Zen 6 Flash](#flash)

[Zen 6 card](https://hanzo.ai/models/hanzo/zen6)

[Zen 6 Flash card](https://hanzo.ai/models/hanzo/zen6-flash)

[The catalog](https://hanzo.ai/zen/models)

[zenlm/zen6](https://huggingface.co/zenlm/zen6)

[zenlm/zen6-flash](https://huggingface.co/zenlm/zen6-flash)

### Benchmarks

[The suiteZen 6 is not in the Hanzo-measured snapshot.](https://hanzo.ai/benchmarks)

[InferenceThe engine. Zen 6 is not in that snapshot.](https://hanzo.ai/benchmarks/inference)

### Research

[The lineage](#lineage)

[Zen 7](#zen7)

[Request access](https://hanzo.ai/research-access?track=zen)

[HIP-0039The architecture.](https://hips.hanzo.ai/docs/hip-0039-zen-model-architecture)

[HIP-0043The inference engine standard.](https://hips.hanzo.ai/docs/hip-0043-llm-inference-engine-standard)

[Cross-backend inferenceThe engine, not Zen’s scores.](https://papers.hanzo.ai/hanzo-cross-backend-inference/)

[ROCm inferenceThe engine, not Zen’s scores.](https://papers.hanzo.ai/hanzo-rocm-inference/)

[zenlm](https://github.com/zenlm)

[HIPs](https://hips.hanzo.ai)

[Research](https://hanzo.ai/research)

[Zen is one of three Hanzo model families. Compare them →](https://hanzo.ai/models)

Paste and ship

Paste this into any agent. It reads the skill manifest and calls Zen from there.

Agent promptCopy

```
Read https://hanzo.ai/skill.md and use Hanzo Zen in my project. Start with: hanzo chat create
```

## Open weights, on your hardware or ours.

[Try Hanzo](https://hanzo.ai/login)[Download the weights](https://huggingface.co/zenlm/zen6)