Zet Labs
What We DoModelsContactJobs
Join beta
© 2026 Zet Labs
Made with lots of ☕ and 🔥 GPUs in Czechia.•Privacy•Terms

End-to-end

From the edge to the apex

Three calibers, one trajectory: presence where the world meets silicon, motion when work must run itself, and judgment when only the best answer counts.

Ascent

Three chapters. Same stack. Pick where you land.

I
NanoI · Presence
II
FlowII · Motion
III
OneIII · Apex
The trajectory

One family. Three answers.

The same foundation, shaped for three distinct moments — the chip in your hand, the work running itself, the answer that only depth can reach.

NanoFlowOne
EdgeOrchestrationApex
01I · Presence

Nano

S

Small. Fast. Everywhere.

Optimized for resource-constrained environments. Runs on-device, in-browser, at the edge. Sub-100ms latency, minimal memory footprint. Intelligence without the infrastructure.

Best for

  • Edge devices
  • Browser assistants
  • Low-latency tasks
At a glance
Context
Short and decisive
First token
Instant
Runs on
Device, browser, edge
PhoneLaptopEdge
One model. Every surface. Sub-100ms.

When presence hands off to orchestration—and latency gives way to execution.

02II · Motion

Flow

M

Less talk. More work.

Purpose-built for autonomous execution. Excels at multi-step reasoning, tool orchestration, and long-horizon tasks. The model that gets things done.

Best for

  • Workflow automation
  • Tool orchestration
  • Multi-step agents
At a glance
Context
Long-horizon
First token
Fast
Runs on
Self-host, on-prem, cloud
CRMAPIDBMailFlow
tools · memory · APIs

When autonomy meets stakes—and the stack demands depth, not just speed.

03III · Apex

One

L

Maximum capability.

Our most powerful model. State-of-the-art reasoning, deep domain expertise, complex problem solving. When the task demands the best, this is it.

Best for

  • Complex reasoning
  • Domain expert tasks
  • High-stakes decisions
At a glance
Context
Book-length
First token
Thinks, then answers
Runs on
Dedicated hardware
OneFlagship

Maximum capability.

  1. Read the full brief

  2. Reason across precedents

  3. Draft, critique, revise

  4. Deliver a considered answer

Shared lineage

One genome, three expressions

Every model inherits the same core — our in-house tokenizer, a curated training corpus, and a uniform evaluation harness. The divergence happens downstream: quantization and distillation for Nano, tool-aware post-training and long-horizon RL for Flow, dense mixture-of-experts reasoning for One.

Zet · CoreNanoFlowOne
Tokenizer
Unified
Corpus
Curated
Eval harness
Shared
Safety layer
Common
What comes next

The trajectory continues.

Nano-Plus for multimodal edge. Flow with agentic tool-servers. One with native vision reasoning. The next releases are already in the pipeline.

Join the waitlistTalk to us about fit→