webAI | Co-Founder & CPO
Austin, TX
Joined February 2022
- Tweets1.1K
- Following366
- Followers392
- Likes1.8K
Tyler Mauer retweeted
Benched Air-26B-A4B on the same 210 tasks:
- Standard 26B Q4_K_M (17.4G): 14.2 tok/s | 180 prefill | 2,400ms TTFT
-❗️26B-A4B (12.55G): 131.9 tok/s (161 pk) | 1,596 prefill | 74ms TTFT | AND top score of all models on the 210 tasks
Tyler Mauer retweeted
Thank you everyone for the support on TwIL-8B.
Seeing how everyone is using the model already is amazing!
Awesome, great work by the team. Incredible work happening at webAI, glad to share some of it publicly.
Today, we’re excited to open-source TwiL-LM3, the first formal reasoning model from the webAI Intelligence Lab.
At just 3 billion parameters, TwiL-LM3 outperforms OpenAI’s GPT-OSS-120B on 4 of 5 formal reasoning benchmarks while running efficiently on consumer hardware. That’s 40× fewer parameters, 2.6× faster inference, and state-of-the-art performance in the reasoning tasks that power reliable tool calling, code generation, structured outputs, and AI agents.
TwiL-LM3 was trained using webAI’s proprietary reasoning pipeline on webAI-owned, verified datasets—not scraped internet data. We believe better reasoning comes from better training pipelines and higher-quality data, not simply larger models. Our approach demonstrates that efficient models can rival—and in many cases surpass—models dozens of times their size.
Designed for the edge, TwiL-LM3 runs on hardware people already own—from a Raspberry Pi to an iPhone—bringing advanced reasoning to millions of devices without relying on the cloud.
This is our first open-source release from the webAI Intelligence Lab, and it’s only the beginning.
Proudly built in Austin, Texas.
Article: webai.com/blog/webai-release…
Awesome project the team put together. Check it out.
We're open-sourcing YOLO26-MLX.
A native MLX implementation of the YOLO26 object detection models, built from the ground up for Apple Silicon. No PyTorch runtime. No external GPU infrastructure. Train and run real-time object detection directly on your Mac.
In internal benchmarks on M4 Pro, the MLX implementation delivered up to 2.6x faster inference and 1.7x faster training compared to PyTorch with MPS. Accuracy stays within 0.5% of the official YOLO26 results.
This has powered object detection inside our products since the YOLOv8 generation. Now it's yours.
Read more: webai.com/blog/running-yolo2…
GitHub: github.com/thewebAI/yolo-mlx
To the next generation, when your calendar starts to feel like the law, being human means taking a break from your programmed time to do what you want.
Tyler Mauer retweeted
Alright listen San Francisco. Everyone cant sell shovels. Someone has to mine something.
How much of "model limitations" are simply misalignment with user preferences?
It is likely that brutal generalization is a large contributor to model limitations.
AI on cloud is Google's game, the internet companies reign supreme in the classic game of resources. Infrastructure disruption is required to create a new wave of high value startups.
Apple should introduce a dedicated utility that repurposes discontinued hardware, such as legacy iPods, into functional external storage devices.