Occamy-1.0: An Open 35B Model for Cost-Efficient Co-Work

Summary

Co-work agents handle workflows that combine information gathering, tool use, coding, and file manipulation across many model calls, so total cost and latency matter alongside peak capability. The authors present Occamy-1.0, a cost-efficient co-work model produced by further training the post-trained Qwen3.6-35B-A3B checkpoint. They build execution-grounded data and environments, collect replayable long-horizon trajectories across multiple harnesses, and apply staged post-training to consolidate complementary execution abilities. Across a broad set of co-work benchmarks, Occamy-1.0 is reported to be among the strongest models of comparable size and remains competitive with substantially larger frontier systems on several tasks. Under the paper's evaluation and pricing protocol, its aggregate results on four representative benchmarks place it at the low-cost knee of the observed cost-performance Pareto frontier. Additional evaluations in tool calling, coding, and instruction following indicate that the specialization retains broad agentic capability. The authors release the model weights and a subset of the training data for research on practical co-work agents and agentic post-training.