← Papers & projects

Model & systems Open model

Occamy-1.0

A 35B-A3B co-work model for long-horizon, tool-using tasks

85.93%
ClawEval strict Pass@3
−38.6%
Tokens per trajectory
−36.1%
Trace wall time

Occamy-1.0 is a 35B-A3B co-work model continued from Qwen3.6-35B-A3B through full-parameter SFT, uniform model soup, and GRPO/SAO reinforcement learning.

Post-training and evaluation

  • A verifier-gated, multi-harness data and training pipeline with immutable provenance and quarantine gates.
  • Token-exact proxy/TiTO replay and environment-state reconstruction for long-horizon tool-use trajectories.
  • Episode-level credit assignment that remains coherent across context rewrites.
  • Frozen, same-harness evaluation for measuring capability and efficiency without silently changing the task contract.

Results

  • Raised Combined ClawEval T/C Strict Pass@1/3 from 65.16/73.87% to 77.39/85.93%.
  • Reduced tokens per trajectory by 38.6% and trace wall time by 36.1%.
  • Reduced timeouts from 10.89% to 4.69% and invalid calls from 3.28% to 1.68%.

Explore the model on Hugging Face →