Optimus-1.2 Released: Let Every Engineering Team Have a Scientific Civilization Running on Its Desktop
Arkenos releases Optimus-1.2, focused on autonomous swarms for long-horizon research. More stable and more token-efficient; some in the community already want to use it to run math experiments.
Dev Shah today announced Optimus-1.2, calling it the first self-improving autonomous swarm tool aimed at long-horizon research and discovery. It will be available to all Arkenos users on Thanksgiving.
Over the past few months, the team focused on solving the hardest problems in autonomous swarms. According to the official account, the swarm is now noticeably more stable, safer, and more token-efficient. The next step is to extract maximum engineering and research value from every token enterprise users spend on Arkenos.
If you are building a frontier engineering or research team, they can help you run thousands of additional experiments. If you are a neocloud or inference platform, they also want to explore how to make your compute do more useful research. The original has a very evocative line: every engineering team should have an entire scientific civilization running on its desktop.
Minutes after the release, the community reacted quickly. Iacon Autonomics replied with 'fascinating,' with a video attached. Eric said he was working on this himself. Daniel White said it looked great.
More interesting is a user named Aditya, who asked whether this feature is included in the existing subscription plan. He built a mathematical language to test and create mathematics in different environments, some of which are non-standard or cannot even be verified. He wanted to throw this problem at Optimus-1.2 to see what it could produce. Dev Shah replied yes and said he had already DM'd him.
A more specific technical discussion came from Brian Narita. He mentioned his MeshForge project, a local-first, declarative requirements agent that currently only has a requirements document and has not implemented anything. Any implementation work will start only after his approval. It connects three projects: Familia, Virtu, and Rakazo; each project publishes a req.yaml declaring interfaces, data contracts, resource limits, consent rules, and security labels, which are then uniformly checked by a verifier and graph solver, conflicts are flagged, and topology and adapter stubs are proposed. For security, it uses a distributed version of the TrustedBSD MAC/LOMAC model. He asked how Optimus-1.2 matches MeshForge's current design requirements and whether there are major pain points. No public reply to this question has been seen yet.
For autonomous swarms to truly land, it still depends on how they collaborate with existing infrastructure. How trust boundaries are drawn will also determine whether it is a demo toy or a daily tool.
发布时间: 2026-11-20 20:40