Extreme speed. Ultra-low cost.
Self-correcting. Parallel generation.

Blue Whale

Diffusion Language Model

Try Model Try Product

Models

First Self-Correcting Foundation LLM

W1-4B-dLLM

Try W1-Instruct W1-Base Research

Latent Reasoning

Variable Length

Parallel Generation · Correction

Same 100 code tokens: LLM needs 100 steps, dLLM only 15 steps

LLM 100 steps
dLLM 15 steps
500+ Effective TPS

A single consumer-grade 4090 breaks 500+ Effective tokens/s

Effective tokens only · not inflated by EOS, padding or space tokens

Products

Agent Matrix

Team

Top-Tier Research Team

Honest · Curious · Bold — with abundant resources, a diverse culture, and a world-class platform.

Whaletech AI team

Educated At

Harvard University
Harvard
Massachusetts Institute of Technology
MIT
New York University
NYU
University of Oxford
Oxford
Imperial College London
Imperial
Australian National University
ANU
The University of Sydney
USYD
Fudan University
Fudan

Partnership

Eyes on the Stars, Feet on the Ground

info@whaletech.ai @whaletech_AI_

Careers

Investors

Partner Companies