Reasoning, evaluation
and learning

Our research spans multi-step reasoning, long-horizon planning and reinforcement learning, with evaluation grounded in verified outcomes

Reasoning paths and evaluationA reasoning graph branches through alternative steps. One teal path connects the starting point to a verified outcome.

Results on public benchmarks

DABstepJuly 2026

Actioneer v0.5

93.78%

Overall accuracy

Hard-set accuracy94.44%
Read benchmark note
DABstep leaderboard
  1. Actioneer v0.593.78%
  2. Nvidia KGMON89.56%
  3. Microsoft 365 Copilot68%
  4. Sphinx AI65.56%
  5. Google DS-Star52%

DABstep by Hugging Face and Adyen: 450 multi-step tasks on real financial data, reported July 2026

KramaBench

Against 66.3% for Claude Code

78.8%

DA Bench

UC Berkeley and PromptQL · Next-ranked system: 65.5%

72%

Research agenda

  1. Know

    Caller classification

    Telling a customer apart from an AI assistant or a synthetic voice within the first seconds of an 8 kHz phone call, across the languages we support

  2. Act

    Planning over weeks

    Planning calls and messages over weeks using data from core banking, policy and pharmacy systems. DABstep measures the multi-step data reasoning this depends on

  3. Measure

    Deterministic and calibrated reward signals

    Scoring calls deterministically where a rule can be checked, and calibrating model judges against human QA reviewers where it can’t

  4. Learn

    Reinforcement learning on verified events

    Post-training open-weight models on verified events such as repaid instalments and collected refills, with each update gated on compliance evaluations

Notes from research
and engineering

All articles

Benchmark an agent on your own rubric

Score sample calls and the account events that followed them with Pulse, then explore where post-training could improve resolution rates

How the platform works