Verne (@VerneRobotics) header
Verne (@VerneRobotics) profile picture

Verne builds general‑purpose physical AI. We deploy robots to expand the productive capacity of the physical world.

Joined June 2025
Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)Photo by Verne (@VerneRobotics)
Introducing V0.5, Verne’s open-weight world action model for industrial robots. We believe democratizing robotics starts with making frontier foundation models open to everyone. Bringing robots to every industry will take many teams, building on shared progress across many workflows. V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks. We hope you’ll join us in putting V0.5 to work.
72
83
24
506
82,881
I visited @VerneRobotics in SF and saw V0.5, their new open-weight world action model, in action. • Full-body tasks from ~2 hrs of data • Beats π0.5 on RoboTwin 2.0 • Two robots coordinating Full video of the tour with @neil_nie_: youtu.be/oj2jw6-HrzI
24
16
11
164
23,730
Verne retweeted
The latest generation of VLMs (GPT-Astra and Fable), has completely changed how I think about robotics. Six months from now, the robot systems we build will look very different from what we’re building today. Thank you for the chat @sophiadew and thanks @mts for having me today!
Verne Robotics CEO @neil_nie_ calls GPT-6 Astra a "watershed moment" for robotics after it learned to control physical robots with very little robotics training data, transferring knowledge from video games and 3D environments into the real world: "The thing that surprised me and the community the most was the fact that it was able to learn how to control a robot without actually being trained on a lot of robotics data." "This is almost saying that the bitter lesson holds true not just in large language models but also in robotics." "This cross-domain knowledge transfer, from 3D control or video games that transfer directly into robotics, was a watershed moment for us." "Without speculating too much on exactly what the model is thinking, ChatGPT has learned a lot about 3D space and moving in 3D. That is very relevant to what it takes to control robots." @VerneRobotics
6
32
2,503
Verne retweeted
Verne Robotics CEO @neil_nie_ calls GPT-6 Astra a "watershed moment" for robotics after it learned to control physical robots with very little robotics training data, transferring knowledge from video games and 3D environments into the real world: "The thing that surprised me and the community the most was the fact that it was able to learn how to control a robot without actually being trained on a lot of robotics data." "This is almost saying that the bitter lesson holds true not just in large language models but also in robotics." "This cross-domain knowledge transfer, from 3D control or video games that transfer directly into robotics, was a watershed moment for us." "Without speculating too much on exactly what the model is thinking, ChatGPT has learned a lot about 3D space and moving in 3D. That is very relevant to what it takes to control robots." @VerneRobotics
Introducing V0.5, Verne’s open-weight world action model for industrial robots. We believe democratizing robotics starts with making frontier foundation models open to everyone. Bringing robots to every industry will take many teams, building on shared progress across many workflows. V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks. We hope you’ll join us in putting V0.5 to work.
3
9
6
52
10,778
Verne retweeted
Very excited to be on the show today! We will be talking about @VerneRobotics, deploying robots, and open source robot foundation models!
1
4
22
2,532
Verne retweeted
LLMs like the newly released Claude Fable 5.2 are making tremendous progress on robotics tasks. There’s a powerful synergy between System II reasoning and planning, and System I fast, responsive execution. V0.5 is designed to work alongside VLMs and LLMs. In this demo for one of our customers, GPT-6 Astra plans how to pack a makeup bag, and V0.5 executes the skills: picking the products, packing them, and zipping the bag closed. As foundation models become better planners, we’re building V0.5 to turn that intelligence into fast, reactive action.
2
1
3
26
1,792
Verne retweeted
Thank you @pentestduck and the @DreamscaleLabs team for becoming the first inference provider to host @VerneRobotics's models! We are working hard to release additional checkpoints on @DreamscaleLabs beyond the benchmark checkpoints. Really excited for the community to try our models!
Verne V0.5 is now live on Dreamscale! You can now use the V0.5 model in LIBERO and RoboTwin environments through Dreamscale's serverless cloud inference.
3
6
2
29
2,056
Run Verne V0.5 today via our API. Thank you for releasing as open-weights @VerneRobotics !
Verne V0.5 is now live on Dreamscale! You can now use the V0.5 model in LIBERO and RoboTwin environments through Dreamscale's serverless cloud inference.
3
5
345
Verne retweeted
Full-body manipulation is essential for industrial work. Robots need to coordinate both arms and a mobile base to reach, pick, and move loads. Many open-weight manipulation models still focus on tabletop tasks. V0.5 handles this full-body grocery handling workflow autonomously, from loading the cart to pulling the cart away.
5
3
38
2,004
Verne retweeted
This embodiment was never seen in pretraining. We collected a small amount of data on this contact-rich assembly task, fine-tuned V0.5, and rolled out the policy in just 1.5 days. V0.5 is an embodiment-agnostic model. If your team is looking to deploy robots, try V0.5! Check out our repo and blog for details: github.com/vernerobotics/v0.… vernerobotics.com/research/v…
7
10
4
102
5,630
Verne retweeted
Open-weight foundation models are critical for deploying robots. Teams deploying robots need the freedom to adapt models to their own hardware, tasks, and environments. Excited to see @VerneRobotics pushing this forward. Congrats @neil_nie_ on the release!
Introducing V0.5, Verne’s open-weight world action model for industrial robots. We believe democratizing robotics starts with making frontier foundation models open to everyone. Bringing robots to every industry will take many teams, building on shared progress across many workflows. V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks. We hope you’ll join us in putting V0.5 to work.
2
3
19
1,653
A team of 11 in San Francisco just released an open-weight world action model with benchmark results beating π0.5, according to Verne. Verne's V0.5 brings that to industrial robots, and the weights are open. That lowers the barrier for every factory and robotics startup out there. Here's what Verne @VerneRobotics just released: • A fine-tuning platform to teach robots new industrial tasks • V0.5, an open-weight world action model • Built for industrial robots • The leading US-made open-weight model on LIBERO and RoboTwin 2.0, according to Verne • Free for research teams to build on Big step for robotics.
Introducing V0.5, Verne’s open-weight world action model for industrial robots. We believe democratizing robotics starts with making frontier foundation models open to everyone. Bringing robots to every industry will take many teams, building on shared progress across many workflows. V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks. We hope you’ll join us in putting V0.5 to work.
28
27
117
6,091
Your robot shouldn’t need better Wi-Fi. Today @VerneRobotics released V0.5, its open-weight robot model. On Change Order, @neil_nie_ explains why Verne runs inference on the robot—and how real deployment constraints shape the product. Full episode ↓
7
6
3
28
4,895
OPEN SOURCE. OPEN WEIGHTS. CAN’T LOSE. new @VerneRobotics release: 🧠 open weights for V0.5, their top US-made OS robot model
 🛠️ a fine-tuning platform to teach tasks in <2 hours
 ✋ DIY grippers to collect your own data need a deep seek moment for US robotics!!! 👇
8
5
1
46
10,970
exciting moment for physical AI, Verne just dropped an open weight model trained on 1.7M demonstrations it’s the top performing open weight model on robotics model benchmarks fine tune with V0.5 OS or let Verne deploy their stack for you 🦾
Introducing V0.5, Verne’s open-weight world action model for industrial robots. We believe democratizing robotics starts with making frontier foundation models open to everyone. Bringing robots to every industry will take many teams, building on shared progress across many workflows. V0.5 is the leading US-made open-weight model on LIBERO and RoboTwin 2.0 benchmarks. We hope you’ll join us in putting V0.5 to work.
1
2
15
1,692
We expect in-context learning to improve rapidly in the near future. Today, we believe few-shot adaptation through fine-tuning is still key to delivering value: collect a small set of demonstrations of your task, then adapt the model to the work you need done.
1
8
730
What will you teach V0.5? Bring us the work you want robots to do. Verne helps you finetune V0.5 on your own demonstrations, and deploy with our team to get it running in production. Put V0.5 to work: vernerobotics.com/signup For more technical details, please visit: vernerobotics.com/research/v… Weights and model can be found on Huggingface: huggingface.co/collections/v… Model and inference code: github.com/vernerobotics/v0.… For commercial partnerships, please reach out to us at [email protected] Let’s give everyone the tools to build the world of their dreams!
12
739