Now in Beta

Introducing the Zylo AI Family of Models

Meet Spark, Orbit and Cosmos — intelligent routers that pick the best model in the world for your task. You choose the tier; Zylo chooses the brain. And on top of the family, two products: Zylo AI Chat and Zylo Code.

The Zylo model family Beta

Zylo isn't a single model. Each Zylo is an intelligent router that, behind one name, selects the best model for your request from a curated per-tier list — and only charges for what the task needs.

Beta

Zylo Spark 1.0

zylo-1.0-spark
The fast, ultra-cheap assistant
Short, structured tasks: classify, extract data, tag, summarize, convert formats. High volume at cents.
$3 / 1M in  ·  $7 / 1M out
Context up to 10M tokens
Ceiling: GPT-5.6 Luna, Gemini 3.1 Flash Lite
Beta

Zylo Orbit 1.0

zylo-1.0-orbit
The balanced workhorse
Your day to day: general chat, RAG, standard coding, tool use and moderate reasoning. The safe default.
$5 / 1M in  ·  $15 / 1M out
Context up to 1M tokens
Ceiling: GLM 5.2, GPT-5.6 Terra, Qwen 3.7 Max
Beta

Zylo Cosmos 1.0

zylo-1.0-cosmos
The expert code reviewer
High-value code review: find bugs, code smells and security issues in code you already have. A reviewer, not a writer.
$15 / 1M in  ·  $30 / 1M out
Context up to 1M tokens
Ceiling: Claude Fable 5, GPT-5.6 Sol, Claude Opus 4.8
Beta access. The Zylo family is rolling out to Go plan and up. Request access from the console and start routing with a single model id — every catalog model stays available under its own name too.

How it works

You pick the tier; Zylo does the rest — live, on every request.

You choose, it selects

Your request lands at the tier you asked for. Inside that tier, Zylo checks live pricing and availability and picks the best model from its curated list — never one outside it.

Sensible escalation

The pricier models in each pool are reserved for the tasks that truly need them. The price you pay per tier stays fixed, no matter which model answered.

Built-in safety net

If a model is down or saturated, Zylo automatically falls back to the next best on the list. Cosmos always keeps a flagship on standby.

Full transparency

Every response includes zylo.routing metadata telling you which tier and model answered, plus an estimated cost.

The Zylo personality, shared by all three tiers:
Asks before actingPlans before executingStructured & rigorousWarm, in your languageFully customizable with your system prompt

What powers each tier

The “base” of each Zylo is the curated set of real models it can choose from. Each tier performs, at its ceiling, like the best model on its list.

TierBase models (curated list)Max context
⚡ SparkGPT-5.6 Luna · Gemini 3.1 Flash Lite · Llama 4 Scout (10M context)up to 10M
⚖️ OrbitGLM 5.2 (primary) · GPT-5.6 Terra · Qwen 3.7 Max · DeepSeek V4up to 1M
🧠 CosmosClaude Fable 5 · GPT-5.6 Sol · Claude Opus 4.8 · Gemini 3.1 Pro (+ omnimodal)up to 1M

Underneath the family sits a catalogue of 69 frontier models from 14 providers, callable directly by their real names (bare ids like claude-opus-4.8, gpt-5.5, gemini-3.1-pro — no vendor prefix).

Anthropic
Claude · Opus / Sonnet / Haiku
OpenAI
GPT-5.x · Codex · GPT-OSS
Google
Gemini 2.5 / 3.x · Gemma
Alibaba
Qwen 3.x · Coder
DeepSeek
V3 · V4 · R1
Z.ai
GLM 4.x / 5.x
Moonshot
Kimi K2.x
MiniMax
M2 · M3
Meta
Llama 3.x / 4 · Scout
Mistral
Devstral · Codestral
xAI
Grok 4.x
Xiaomi
MiMo V2 / V2.5

Capabilities & benchmarks

What each tier can do, and the “capability ceiling” of the best model on its list.

Capability⚡ Spark⚖️ Orbit🧠 Cosmos
Text
Vision (images)
Tool / function calling
Audio / video✓ (omnimodal)
Max contextup to 10Mup to 1Mup to 1M
Codebasicstrong (write + review)review only (bugs, security)
Deep reasoningmoderatemaximum
Streaming
Capability ceilings — headline scores from public benchmarks and vendors (2026), continuously validated with our own evals: Spark — GPT-OSS 120B ≈ 90% MMLU. Orbit — GLM 5.2 is a top open-weight model; DeepSeek V4 ≈ 79% SWE-bench Verified. Cosmos — Claude Fable 5 ≈ 80.3% SWE-bench Pro; Claude Opus 4.8 ≈ 88.6% SWE-bench Verified & #1 on LMArena; Gemini 3.1 Pro ≈ 94% GPQA.

Pricing

Each Zylo tier has a fixed price — it doesn't vary with which internal model answered.

ModelInput / 1MOutput / 1M≈ 1,000 requests*
⚡ Zylo Spark 1.0$3$7≈ $4.50
⚖️ Zylo Orbit 1.0$5$15≈ $8.50
🧠 Zylo Cosmos 1.0$15$30≈ $21.00
Recommended mix — 70% Spark / 20% Orbit / 10% Cosmos≈ $6.95
How billing works. Usage is billed from your prepaid credit balance at each rate — there is no markup on usage; Zylo's flat 25% platform fee applies only when you add credits. Your plan (Basic $0 · Go $10 · Pro $50 · Mega $200 · Enterprise $400 per month) sets your daily limits and includes starting credits. The Zylo family is available from Go and up. *Average request ≈ 800 input + 300 output tokens.

Try it in 2 minutes

100% OpenAI-compatible. Already using the OpenAI SDK? Change two lines — base_url and your key — and set the model to a Zylo tier or any catalog id.

Terminal
curl https://api.zyloai.net/v1/chat/completions \
  -H "Authorization: Bearer YOUR_ZYLO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zylo-1.0-orbit",
    "messages": [{"role": "user", "content": "Hello from Zylo!"}]
  }'
Python
# pip install openai
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_ZYLO_KEY",
    base_url="https://api.zyloai.net/v1",
)

response = client.chat.completions.create(
    model="zylo-1.0-orbit",   # or "claude-opus-4.8", "gpt-5.5", ...
    messages=[{"role": "user", "content": "Hello from Zylo!"}],
)
print(response.choices[0].message.content)
Node.js
// npm install openai
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "YOUR_ZYLO_KEY",
  baseURL: "https://api.zyloai.net/v1",
});

const response = await client.chat.completions.create({
  model: "zylo-1.0-cosmos",
  messages: [{ role: "user", content: "Review this diff for bugs..." }],
});
console.log(response.choices[0].message.content);

Useful endpoints: POST /v1/chat/completions · POST /v1/zylo/chat · GET /v1/models · GET /v1/stats. Works in Cursor, Cline, aider, Continue and Roo Code.

Products built on the family

Two ways to use the Zylo family without writing a single line of glue code.

💬

Zylo AI Chat

A premium AI chat that does far more than talk — it runs code, browses the web for you, talks by voice, builds custom agents and connects to your tools.

  • Canvas: live code editor + preview, console and one-click ZIP export.
  • Zylo Browser: a real cloud browser the agent drives, with live view and recording.
  • Zylos: custom agents with their own model, tools, knowledge base and MCP servers.
  • Live voice: real-time voice sessions with screen & camera sharing.
  • Deep Research: exhaustive investigation across hundreds of sources.
  • Integrations: Google Workspace, Slack, GitHub, Fireflies, YouTube, MCP.
Real use case: an analyst runs Deep Research across 200+ sources, gets a synthesized report in tables, and exports it to Google Docs — in one conversation.
🧱

Zylo Code

An in-browser AI app builder. Describe what you want; an Architect agent plans it, writes real files, previews it live, tests it, pushes to GitHub and deploys — no local toolchain.

  • Architect: plans, asks, scaffolds, tests and ships in one agentic loop.
  • Live preview: your app renders as it's written, character by character.
  • Build modes: Plan, Fast, Orchestrator (5 specialists), Security, Custom.
  • 1-click deploy: to Vercel, Cloudflare, Netlify, Railway or Cloud Run.
  • GitHub: auto-save every build, push and open pull requests.
  • BYOK & design tools: bring your own provider + an 88-component catalog.
Real use case: from a one-line idea to a working React app with live preview, a GitHub repo and a live deployment — in an afternoon, with no local setup.

Frequently asked questions

What exactly is the Zylo model family?

Spark, Orbit and Cosmos are intelligent routers. Behind one model id, each picks the best model for your task from a curated per-tier list — so you get quality without choosing a model on every request.

Can I force a specific tier?

Yes — the model id is the tier. zylo-1.0-spark will never spend like Cosmos. And you can always call any catalog model directly by its own id.

Can I see which model answered?

Yes. Check zylo.routing.model in the response, along with the tier and an estimated cost.

Does it support streaming and tools?

Yes — set "stream": true like any OpenAI-compatible model, and pass your tools as usual. All three tiers support vision and function calling.

Do I need a paid plan?

The Zylo family is in Beta and rolls out from the Go plan and up. The free Basic plan still gives you Basic-tier catalog models with a daily allowance.

Start building with Zylo

Create a free account, grab an API key, and route to the best model with one line — no credit card to begin.

Get your free API key