Hedra brings the self-improvement loop behind frontier research to robotics, manufacturing, biology and defense teams. In under a week, its team made a production image model roughly 6x faster on the same hardware.

San Francisco, CA, Nov. 10, 2026 (GLOBE NEWSWIRE) -- Hedra, an a16z-backed research company building models and infrastructure for multimodal and physical AI, today launched Hedra Model Factory, an autonomous post-training and inference platform for building models for the real world. The platform combines flexible GPU sandboxes, post-training run by Hedra researchers working with autonomous AI agents, co-optimized low-latency deployment, and a data flywheel that keeps improving each model after it ships.


Michael Lingelbach, founder and CEO of Hedra.

Models that improve themselves are accelerating frontier research. Hedra Model Factory applies that loop to real-world problems, where data is multimodal and orders of magnitude larger than text, and where reinforcement learning environments and simulators often need GPU acceleration just to run at scale. Hedra grew revenue 15x over the past 12 months, and the platform opens the research and infrastructure behind that growth to teams that need custom models built on their own data.

“Recursive self-improvement is already changing how frontier models get built,” said Michael Lingelbach, founder and CEO of Hedra. “The next step is the real world, where the data is video, sensors and simulation. Hedra Model Factory gives robotics, manufacturing, biology and defense teams the infrastructure to post-train, deploy and keep improving their own models.”


From Compute to a Data Flywheel
Compute. Hedra GPU sandboxes give teams flexible capacity running on Hedra’s inference engine, so they can scale training up without locking into long-term GPU commitments.

Post-training. Teams bring their simulation engines and data, and Hedra helps build the reinforcement learning environments and datasets they need. Hedra researchers work alongside each customer’s team, using tools such as vision-language models that caption and score data or give feedback during online reinforcement learning. For companies without their own research lab, Hedra researchers also set up data capture, evaluations and the training-to-serving loop.

Deploy. Models run on low-latency compute, on premises or at Hedra's nearest partner data center, from distilling world action models to run faster than real time without losing physical coherence to fitting a vision-language-action model onto an edge device. Teams can also bring an existing model to serve on Hedra’s inference engine.

Self-improve. Once a model is deployed, Hedra researchers analyze its production performance, retrain the model and release only versions that pass the customer's evals.


Research Results Behind the Factory
Physically consistent video. Hedra retrained H3 for physical consistency. The retrained model scored 34.9 on Google DeepMind’s Physics-IQ benchmark, compared with 30.3 for H3 Max, with comparable visual quality. An autonomously distilled version rendered a five-second clip in under three seconds on AMD MI355X GPUs.

Image editing. Through distillation and inference optimization, Hedra cut 2K image edits from about 22 seconds to 3.6 seconds on the same four AMD Instinct MI355X GPUs, roughly 6x faster, in under a week. The baseline ran the original bf16 model at 30 sampling steps; the faster result uses a distilled 14-step model. The model runs on MI355X on Hedra today.

Robot hardware changes. In a simulation study, Hedra fitted a robot running a trained policy with a new gripper and restored task success from near zero to near 100% with no retraining, by matching the new gripper’s reach and grasp point in the wrist camera view.

Recursive self-improvement describes systems that get better at improving themselves. Hedra applies the idea to how models are made: researchers work alongside autonomous AI scientists that run experiments and iterate on training runs, and every deployed model feeds data back into the next version. Hedra’s research pushes toward a model factory that gets faster with every model it produces.

Lingelbach has spent his career on reinforcement learning for visual and physical systems, first at Stanford and NVIDIA. Hedra’s research team brings together talent from Stanford, Berkeley and UIUC, with experience at NVIDIA, Meta Superintelligence Labs, Fireworks AI and Salesforce, and has published research in generative video and embodied AI at NeurIPS, CVPR and ICLR.

Availability
Hedra Model Factory is available starting today at the new hedra.com. Teams can work with Hedra researchers to build and deploy a custom model for the problem their general-purpose model can't solve. They can also start with flexible GPU sandbox capacity without long-term commitments or serve an existing model on Hedra's inference engine.

Media images are here. 

About Hedra
Hedra is an a16z-backed research company building models and infrastructure for multimodal and physical AI. Learn more at hedra.com.

CONTACT: For more information, please contact Everett Butler at Hedra on everett@hedra.com