Context
In NapsixB2B the work is done by XIA, and behind XIA there is a model that decides which tool to use, in what order and with which data. That model is now ours: Mato v1.0. Mato v1.0 was trained on Alibaba Cloud's Qwen family (Qwen 3.8 27B base, Apache-2.0 license, official FP8 quantization) and is now published openly on Hugging Face: anyone can view it, accept the terms and clone it to build on top. Want the full experience? Our infrastructure serves it free over the API.
What changes
- Verified tool-calling on Napsix tools. Mato doesn't *mention
- your apps: it operates them. It reads and writes spreadsheets, searches and updates contacts and deals in the CRM, reads mail threads, crawls the web and drafts documents.
- 26 native apps and 24 chat commands available from the first message, with no connectors to build and no keys to configure.
- It plans before it executes. It picks the tools it needs and the order to use them before touching anything, then delivers the artifact ready to review: the sheet, the contact, the report.
- Long context: 256k native tokens, extendable to 1M with rope scaling. Flexible thinking, on by default in direct mode.
Spec sheet
| Specification | Mato v1.0 |
|---|---|
| Context | 256k native · up to 1M with rope scaling |
| Input | Text · vision (image and video): being validated |
| Tool-calling | Verified on Napsix tools |
| Thinking | Flexible, on by default in direct mode |
| Price | $0 per token, on every plan |
| Served on | NVIDIA H100 80GB, our own infrastructure |
Production performance
- ~2s for short answers
- ~30 tok/s streaming
- 580k KV cache tokens
- 17.7× concurrency
Measured on Daytona H100; varies with load.
What it means for you
- Zero cost per task. Mato is served at $0 per token on every plan. Combined with routing each task to the right-sized model, every job can go to the cheapest engine that solves it well.
- Less integration surface. Where a generic model needs connectors someone has to build and maintain, Mato goes through the system's native tools: Gmail, Sheets, HubSpot and the rest of your integrations.
- No lock-in. Apache-2.0 license and open weights: you can audit the model and, if your data policy requires it, clone it and self-host.
Generic model vs. Mato on Napsix tasks
| Criterion | Generic model | Mato v1.0 |
|---|---|---|
| Access to your data | Needs connectors you build | Napsix tools natively |
| Tool calls | Manual or nonexistent | Verified on Forms, Sheets and CRM |
| Context per task | One model for everything | Routed to the right size |
| Cost | Paid per token | $0 per token, on every plan |
What we're doing
Two versions are already in training: Mato 1M, with 1M tokens of context to fit whole projects, repos and datasets into a single conversation, and Mato Flash 1M, the fast variant with the same context, built for volume, classification and parallel agents. In the meantime, Mato v1.0 is live: try it free over the API or clone it from Hugging Face.
Source: Mato v1.0 — napsix.com.
