Post

Conversation

Two 23 year old Indians just dropped the #2 open-weight AI voice model in the world, trained purely on free credits! Maya1 is #20 globally, better than even Google's best. 3B params, runs on one GPU and does 20+ emotions with < 100ms latency You can just do things.

0:13

8:39 PM · Nov 4, 2025

11.9K

Views

Post your reply

Deedy

@deedydas

41m

Source: huggingface.co/maya-research/ Thanks

@Dheemanthredy

and Bharat for showing the world you can in fact train some awesome AI models very cheaply if you dare to dream

maya-research/maya1 · Hugging Face

Awesome! I was hoping to see sonic-3 by

too on this list?

Sonic 3 is #22

Text to Speech Leaderboard | Artificial Analysis

From artificialanalysis.ai

630

SaaS Junction || AI & SaaS Updates

@SaasJunctionHQ

38m

3B params on ONE GPU is actually insane. Most models this capable need clusters. Maya Research basically democratized voice AI for small studios and indie devs. Game changer for regional content creators.

The hunger in India based talent is real. India is lagging behind in AI innovation now. But it will quickly catch up.

google’s next “breakthrough” will just be a rebranded version of this model with 100x the compute budget

Politics and corrupt bureaucracy is the root cause in India for not innovating, most tech Indians are just waiting to shift to SF, like you.

Love to see it!

lfg

Dope

Incredible. Free credits to world-class model better than Google. This is the maximalist compute thesis in action. When compute is this cheap, the resulting output explosion is inevitable. The floodgates are opening.