Open weights · run anywhere · upstream-reported benchmarks

The Zen family

55 open-weight models across language, code, vision, image, audio, and retrieval — co-designed by Hanzo AI and the Zoo Labs Foundation. Free to self-host, or managed on Hanzo Cloud. Benchmarks here are UPSTREAM-reported for the open ecosystem Zen builds on; only Enso is Hanzo-measured end-to-end.

Four generations, one API

MoDE (Mixture of Diverse Experts) architecture. Explore the full catalog with specs and pricing.

The open-weight landscape — upstream reported

Where the open ecosystem Zen builds on stands, by benchmark. 44 open-weight models; numbers are upstream-reported unless tagged Hanzo — we only relabel a score as ours when we ran it. Toggle the provenance to see which is which.

ModelGPQA-DiamondSource$/MTok out
kimi-k2.689.1Vals AI$2.71
qwen3.5-397b-a17b88.4LLM Stats$2.04
nemotron-3-ultra-550b-a55b86.1Vals AI$1.54
glm-5.285.6Vals AI$3.73
glm-5.184.5Vals AI$3.63
kimi-k2.584.1Vals AI$1.69
glm-583.3Vals AI$2.07
nemotron-3-super82.7LLM Stats$0.4
minimax-m2.582.1Vals AI$0.76
mimo-v2.581.6Vals AI$0.24
deepseek-v3.280.3Vals AI
deepseek-v3.2-exp79.9DeepSeek-V3.2-Exp model …
deepseek-v4-pro76.3Hanzo$2.5
deepseek-r171.5DeepSeek-R1 model card (…
deepseek-4-flash70.7Hanzo$0.2
llama-4-maverick69.4Vals AI$0.75
llama-3.3-70b-instruct50.5LLM Stats

Vendors report on their own harness; Hanzo measures everyone on one. Where both exist the gap is the harness talking — not the model getting better. Hover a source for its exact provenance.

Why upstream-reported? Zen is the open-weight family; its results follow the reported numbers of the open bases it is built on, on the vendors’ own harnesses. Enso — the proprietary orchestration layer — is the one thing we measure end-to-end on a single common harness. That is the honest line between what we ran and what we cite.