Hugging Face has introduced Occamy-1.0, a cost-efficient co-work agent model that the team describes as achieving strong performance on complex multi-step tasks while retaining broad agentic capabilities at lower cost. The model is obtained by further training the post-trained Qwen3.6-35B-A3B checkpoint.
The paper frames co-work agents as systems that execute complex workflows combining information gathering, tool use, coding, and file manipulation across many model invocations. Because cost and latency accumulate over an entire episode, the authors argue that practical value depends not only on peak capability but also on how efficiently that capability is delivered. They note that many steps in everyday work emphasize state tracking, coordination, recovery, and follow-through rather than frontier-scale reasoning.
To build Occamy-1.0, the team constructed execution-grounded data and environments, captured replayable long-horizon trajectories across multiple harnesses, and applied staged post-training to develop and consolidate complementary execution capabilities.
Across a broad suite of co-work benchmarks, Occamy-1.0 is reported to be consistently among the strongest comparably sized models and remains competitive with substantially larger frontier systems on several tasks. Under the stated evaluation and pricing protocol, its aggregate performance across four representative benchmarks places it at the low-cost knee of the observed cost-performance Pareto frontier. Supporting evaluations in tool calling, coding, and instruction following further indicate that the specialization preserves broad agentic capability.
The authors emphasize that for agents, this tradeoff matters considerably, since a single task can invoke the model hundreds of times, making intelligence per dollar as important as intelligence per call. The model weights and a subset of the training data are being released to support research on practical co-work agents and agentic post-training.