The archive · AI & Models · Product decision · 2025–2026
Stellon Labs bets sub-25MB CPU-only TTS can close the gap with cloud voice
YC S25 research lab trains tiny TTS models under 25MB that run anywhere with no GPU — 8K GitHub stars, 45K downloads, 561 HN points.
Stellon Labs
What the business is
Stellon Labs trains 'tiny frontier' AI models for speech, language and video that run on smartphones, wearables, robots and embedded systems; first product is Kitten TTS, an open-source ONNX text-to-speech library (15M–80M parameters, 25–80MB) with commercial support and enterprise licensing.
Starting capital：Not disclosed; YC Summer 2025-backed with a 2-person team per the YC company page.
How it started
Founder Rohan Joshi (HN author rohan_joshi) started Stellon Labs in 2025 on the observation that on-device AI is bottlenecked by a lack of tiny models that actually perform. The 2-person lab went through YC Summer 2025 in San Francisco and shipped Kitten TTS — a super-tiny, CPU-only text-to-speech model under 25MB — as its first open-source product.
What happened
In March 2026 the lab released three new Kitten TTS models — 80M, 40M and 14M parameters — with the smallest under 25MB, claiming SOTA expressivity among similar-sized models, and eight voices (four male, four female). Quantized int8+fp16 and ONNX-based, they run on Raspberry Pi, low-end smartphones, wearables and browsers with no GPU. The Show HN on 2026-03-19 drew 561 points and 181 comments; the company later added a free hosted API and keeps an optimized engine, mobile SDK, multilingual TTS and KittenASR on its roadmap.
How it ended up
Still live: as of September 2026 the repo is in developer preview with a 15M–80M model family, Apache 2.0 license, free hosted API, and commercial support and enterprise licensing via info@stellonlabs.com; YC lists an active 2-person San Francisco team.
Background
Stellon Labs is a YC Summer 2025 research lab betting that edge AI's bottleneck is a lack of tiny models that actually perform. Its founding argument: existing foundation models need so much compute and memory that they are inaccessible on smartphones, wearables, robots and embedded systems.
Its first product, Kitten TTS, is an open-source, CPU-only text-to-speech library under 25MB. In March 2026 the lab released three new models (80M, 40M and 14M parameters), with the 14M variant claiming SOTA expressivity among similar-sized models, eight voices, and ONNX int8+fp16 quantization designed to run on Raspberry Pi, low-end phones, wearables and browsers — no GPU required.
The Show HN on 2026-03-19 hit the front page with 561 points and 181 comments; YC reports 8K GitHub stars and 45K model downloads within two weeks of launch. As of September 2026 the repo is in developer preview with a free hosted API, and the lab monetizes through commercial support, custom voices and enterprise licensing.
What has to be true
- Reversed the scaling orthodoxy: instead of frontier-size models, Stellon trained the smallest model that could still sound good, making 'no GPU required' the entire pitch.
- Open source as distribution: Apache 2.0 plus 8K stars and 45K downloads in two weeks converted developer attention into enterprise licensing leads.
- A narrow wedge with a roadmap: TTS first, then engine, mobile SDK, multilingual models and ASR — each step deeper into the same edge-AI bet.
- YC Summer 2025 validation let a 2-person lab sell commercial support to software and hardware companies that want to run AI locally.
What can be applied
Ship the smallest model that wins the job: a sub-25MB TTS that sounds good makes 'no GPU required' the pitch, and open source plus commercial support turns HN attention into enterprise leads.
Aftermath
As of 2026-09-02 Stellon Labs is live and active: Kitten TTS is in developer preview with models from 15M to 80M parameters, a free hosted API at platform.kittenml.com, eight built-in voices, Apache 2.0 licensing, and commercial support for integration, custom voices and enterprise licensing via info@stellonlabs.com. The roadmap lists an optimized inference engine, mobile SDK, higher-quality and multilingual models, and KittenASR. YC's directory shows an active 2-person San Francisco team; funding amounts are not disclosed.
Sources
- Show HN: Three new Kitten TTS models – smallest less than 25MB
- KittenML/KittenTTS
- Stellon Labs: Building tiny frontier AI models that run on edge devices
- Launch YC: Stellon Labs: Building Tiny AI Models for Edge Devices
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card