How Mubit launched on X
Reliable agents. A model for every task.
What Mubit launched

We started Mubit in Aberdeen. No network, no lab, nobody knew us. We just wanted AI systems that actually work. Today we work with some of the best companies and engineers in the world. And the thing we kept seeing: you and your competitor run the same AI. It should be yours. https://t.co/lIQIfN5JJx
How TLVC rated the launch
TLVC Rating
About the launch
Mubit is for teams whose agents already call OpenAI, Anthropic, or Gemini and repeat the same jobs. Change the client base URL, keep your model IDs, and calls continue on your provider key while Mubit records them. Calls that share instructions, tools, tool choice, output format, and model become one task. From kept examples, Mubit trains a small adapter for that task and tests each version against the provider model's answers. It refreshes a dedicated model for repeated work and switches that task over only when the dedicated model does better on those tests. Frontier models stay available, and models built from a team's own calls stay with that team. Traces and verified outcomes can also become reusable memory and candidate changes to prompts and skills, which you review before activation. Anil Peri, co-founder and CEO, started Mubit in Aberdeen in 2025. This launch is the product answer to a pattern he describes in the announcement: a company and its competitor often run the same rented models, while the repeated work one side already does can become models that side owns. The release covers the platform, Banyan Sapling, and latte/bench. Banyan Sapling, the first model in the Banyan family, is an Apache 2.0 research preview with public weights. Given a policy, evidence, and a claim, it labels the claim supported, a violation, missing information, or inconsistent context, and can return unsure when calibration does not support a unique decision. In the published evaluation of 1,731 cases, 779 received a calibrated answer and all 779 matched the reference labels. The other 952 stayed unresolved. latte/bench checks whether measured feedback changes the next plan. A model specifies 33 bounded parameters for a simulated latte pour, a fixed executor runs that plan, and the model may try again with a short note from the score. Saved notes raised the average best-of-five score for seven of eleven models with valid plans and lowered it for four. Teams that already store calls in Langfuse, Braintrust, or Pydantic Logfire can have an admin import up to 30 days of history with a read-only key, so Mubit can find tasks in work that has already run.
Tags
Questions
Frequently asked questions
The Mubit launch post reached 3.2M views on X, with 2.0K likes, 195 reposts, 205 comments, 710 bookmarks.
Mubit launched on X on October 23, 2026 with the message "Reliable agents. A model for every task.".
Mubit is for teams whose agents already call OpenAI, Anthropic, or Gemini and repeat the same jobs.
Launchvideo.com rated the Mubit launch 3.20 out of 5: hook 3/5, editing / creativity 4/5, copy 3/5, sentiment of launch 3/5, distribution strategy 3/5.







