Hedra launches Model Factory, an autonomous post-training and inference platform for real-world AI. November 10, 2026 by BPM Team 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 - November 10, 2026; 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. 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. Krea 3 image editing. Through distillation and inference optimization, Hedra cut Krea 3 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. Krea 3 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. About hedra. Hedra is an a16z-backed research company building models and infrastructure for multimodal and physical AI. Learn more at hedra.com.

Hedra's MCP and CLI: Bring Hedra inference into your own agents. Avia Haimovich · August 31, 2026 Hedra is an inference and research company building the infrastructure for visual intelligence. Its goal is to make ubiquitous visual intelligence possible. As part of that, Hedra Inc. is expanding its product surface with the Hedra CLI and MCP, built for bringing Hedra into your own agents and your own automations. Whatever you already use to get work done, ChatGPT, Claude Code, Cursor, Codex, Hermes, or Pi, can now reach Hedra's full inference library directly: accelerated open-source and third-party image and video models, with large language models coming soon, all served on its own accelerated inference stack. You don't need to open Hedra to generate with Hedra, ask the agent you're already talking to, and the CLI or MCP handles it, saved to your Hedra library through MCP, or as a file you download through the CLI. Agents have already made teams more productive everywhere else. This brings that same leverage to visual inference. Automate entire functions, generate personalized sales videos and demo content on the fly, or hand your whole ad and marketing pipeline to an autonomous agent built on Hedra. Your team's agent already knows how to do this. Most teams don't need to be sold on using Claude or ChatGPT anymore, that call's already been made. What's still clunky is everything downstream of it: someone needs an image or a video, so they close the chat, open a different tab, and go make it somewhere else. MCP closes that gap. Connect Hedra once, and generating a visual becomes one more thing the agent your team already runs everything through can do, not a reason to leave it. For an enterprise team, that matters more than it sounds like it should. Rolling out a new tool to a hundred people means procurement, a training session, another login to manage. Rolling out an MCP connector to a team already on Claude or ChatGPT is closer to flipping a switch: everyone who already has agent access now has Hedra access, without anyone individually deciding to adopt something new. What is an MCP server? MCP stands for Model Context Protocol. It's a standard that lets an agent like Claude or ChatGPT connect to outside tools and use them directly, inside the same conversation, instead of just answering questions about them. Hedra publishes an MCP server so an agent can generate images and video the same way it already reads a file or searches the web, one more tool it has access to. What is a CLI? CLI stands for command-line interface, a program you or an agent runs directly in a terminal instead of through a chat window. Hedra's CLI works like any other command-line tool: install it, sign in, and run a command like hedra-cli queue submit to generate something and get a file back. MCP or CLI? Both are legitimate ways to generate with Hedra, and most people end up reaching for each one at different times, depending on what they're doing and what's their workflow. | / | MCP | CLI | | Where the output goes | Saved to your Hedra library, plus a link to the file | A link to download the file, nothing saved to your library | | Setup | Connect it like any other MCP server, no install | Install via npm, Homebrew, or a script | | Billing | Credits, or pay-as-you-go with autobilling | Credits, or pay-as-you-go with autobilling | What you can build. * Keep the context you already built. Brainstorm an idea with Claude, and by the time you're ready for a visual, the conversation already holds the whole thing: the direction, the references, why you talked yourself out of the first two versions. Copy all of that into Hedra's canvas and you're starting over. Ask for the visual right there instead, and it generates from everything you already said, not a prompt you have to reconstruct from memory. * Bring in the design files you already have. If Figma's MCP, or whatever else your team runs, is connected to the same agent, that context travels with it. Ask for a visual that matches an existing brand file, and the agent already has it, instead of you exporting it and re-uploading it somewhere else. * Swap a visual without leaving the chat. Ask Claude to change the hero image on your homepage, and get back the new version, no design queue, no separate tab. * Turn a site refresh into one script. Point the CLI at every image on your site and regenerate the batch overnight, instead of opening a canvas hundreds of times. * Keep your company's channel stocked without booking a shoot. Point the same agent that already runs your content calendar at Hedra, and a new video becomes one more thing it generates on schedule. * Generate a personalized sales video on demand. Demo content and sales videos become something your agent can produce on the fly, instead of a separate production request to someone else's queue. * Let the pipeline run itself. Hand your ad and marketing pipeline to an agent that already plans it, generation included, instead of handing visuals off as a separate step. Getting started. Pick whichever fits how you work. Using the CLI: * Install it, npm install -global @hedra/cli, Homebrew (brew install hedra-labs/tap/hedra-cli), or the installer script at hedra.com/develop. Source is on GitHub if you'd rather build it yourself. * Sign in, two ways to do it: * With a key: grab an API key from your Hedra account and set it as HEDRA_API_KEY. * Without a key: run hedra-cli auth login and authenticate in the browser instead. No subscription needed either way. * Run a command yourself, or point a coding agent at it: hedra-cli jobs submit -model minimax-h3 -input '{"prompt": "a fox sprinting across fresh snow"}'. Using MCP: * Connect it the same way you'd add any other MCP server to Claude or ChatGPT. Full setup steps live at hedra.com/mcp. Either way, pay with your existing credits or set up pay-as-you-go with autobilling, no subscription required. Try it. You already have an agent, it just didn't have Hedra yet. Connect the MCP at hedra.com/mcp, or install the CLI at hedra.com/develop, whichever fits how you work. The next image or video you need is one message away, not a separate tool to open. Faq. Do I need a Hedra account to use the CLI or MCP? Yes, either way, but it's free, no subscription needed. MCP works inside the account you already have. The CLI just needs you signed in too, even though it doesn't save anything back into your library. Do I need to set this up separately for everyone on my team? No, if your team already rolls out Claude or ChatGPT centrally, Hedra's MCP works the same way. Connect it once at the workspace level, and everyone with existing agent access has Hedra access too, nobody sets up their own separate connection. Does my CLI generation show up in my Hedra account? No. You get a link to download the file. MCP generations land in your library automatically; CLI ones don't. Is this the same as Hedra's API? Not exactly. The API and SDKs are for building your own product on top of Hedra's models, for other people to use. MCP and CLI are for reaching your own account, or getting a file back, through an agent you already use, not for building something separate. Which models can I reach through this? The same library the API serves: MiniMax H3, Veo, Nano Banana, Seedance 2.5, and the rest of its accelerated image and video catalog, plus its own in-house models. Large language models are coming soon. Anything Hedra Inc. serve, your agent can reach. Which agents work with this? Claude, ChatGPT, Cursor, Claude Code, Codex, Hermes, and Pi are all confirmed to work, through whichever surface fits how you're working. Do I need a subscription? No. Pay with your existing credits, or set up pay-as-you-go with autobilling.