Just hours after Anthropic dropped a new Claude Code dynamic workflow and ultracode mode alongside Opus 4.8, the team behind open-source framework OMO fired back on X, publicly accusing Anthropic of copying its multi-model orchestration architecture "pixel for pixel."
https://twitter.com/q_yeon_gyu_kim/status/2060215394334552103
OMO — the officially recognized No. 1 plugin for OpenCode, a 167,000-star open-source terminal agent project — was built by 23-year-old Korean hacker Q and is promoted by the operations team Sisyphus Labs. It has racked up 60,000 stars and serious clout in the developer community.
The blowup lays bare Big Tech's "block first, absorb later" playbook. In January, Q introduced ultrawork, a multi-model dynamic orchestration workflow in OMO, along with a coordination brain called atlas that handles merge reviews. Because the orchestration mechanism devours a massive number of tokens when running multi-agent recursive loops, OpenCode co-founder Dax Raad called it a "token burner." The high token consumption directly prompted Anthropic to crack down in April, cutting off third-party clients from using subscription credits.
Just three months later, Anthropic officially absorbed that open-source-vetted dynamic orchestration logic, turning it into a closed-source, paid feature inside its own Claude Code.
OMO's beef isn't just with Anthropic. The team has a long-standing feud with FactoryAI going back to February. FactoryAI's hyped Missions multi-day autonomous task flow was accused of directly ripping off OMO's three-layer agent architecture: OMO's planner Prometheus became Factory's plan approval flow, the coordination brain Atlas was repackaged as orchestrator Droid, the executor Sisyphus Junior turned into Worker Droid, and even the original skill filtering mechanism was simplified and reused.
Despite OMO sharing its historical development records and demanding that the big companies publicly acknowledge its original contributions, both Anthropic and FactoryAI have stayed silent. The open-source team continues to fight back on multiple fronts on X.
On one front, Sisyphus Labs emphasizes that OMO is completely model-agnostic — users don't need expensive official quotas. They can run the ultrawork workflow with local or cheap models like GPT, Gemini, or even Kimi. The team also reveals it's working on a Codex-compatible version based on GPT-5.5.
On another front, Q personally tested the new model and delivered a scathing review, calling it "just a decent midweight GPT-5.5 model." He also fired back with a humorous jab: he used Anthropic's own new tool to command it to rewrite OMO's original orchestration logic, turning it into a compatibility plugin that can be reverse-mounted onto Claude Code. Q joked that he's just helping Big Tech run a compatibility experiment — after all, everyone's used to open-source ideas being taken directly.