Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work

Occamy-1.0, an open 35B co-work agent model trained from Qwen3.6-35B-A3B, lands at the low-cost knee of the cost-performance Pareto frontier across four representative co-work benchmarks, according to its arXiv paper. The model is positioned to let production agent stacks route long-horizon workflow steps — tool use, coding, file manipulation, recovery, and coordination — to a cheaper specialized model instead of frontier-scale systems.

arXiv https://arxiv.org/abs/2609.11977 Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. Occamy-1.0 is an open 35B co-work agent model trained from Qwen3.6-35B-A3B that lands at the low-cost knee of the cost-performance Pareto frontier across four representative co-work benchmarks. For production agent stacks, this means many long-horizon workflow steps—tool use, coding, file manipulation, recovery, and coordination—can be routed to a cheaper specialized model without defaulting every invocation to frontier-scale systems. Occamy-1.0 delivers frontier-class agentic and tool-use performance on a 35B open-weight footprint, establishing a new low-cost Pareto frontier for multi-step workflows. By training specifically on long-horizon, execution-grounded trajectories, this model enables production teams to migrate complex agent loops—such as state tracking, tool calling, and error recovery—from expensive proprietary APIs to cost-effective self-hosted hardware. This drastically slashes the cumulative compounding costs of multi-turn agent episodes without sacrificing execution reliability.