Technology / WORLD

Researchers outline a multi-session workflow for Olympiad agents

HOA combines independent review, structured knowledge notes and formal verification.

Reading sizeSources ↓Share story ↗
Conceptual illustration of technology, processors and connected information.Illustration
Symbolic illustration

Researchers introduced HOA, an open workflow that coordinates multiple agent sessions on difficult mathematics and science problems. It sends candidate solutions to an independent reviewer, supplies missing domain knowledge through a structured wiki, and checks Lean-formalized proofs with a trusted verifier.

The authors report solving 672 PutnamBench problems and 251 problems on Lean-Eval with HOA. Experiments also cover the official 2026 mathematics, physics, chemistry and programming Olympiad problem sets; scoring varies by field and uses human review, code tests or an LLM judge. The findings are limited to these tasks and evaluation procedures and do not establish general scientific-discovery ability.

Data

Table

Measure

PutnamBench result in the paper's evaluation

PutnamBench problems solved with HOA672PutnamBench result in the paper's evaluation
Methodology

Values come from the paper's stated benchmark experiments; they are not field measurements or general performance guarantees.

Updated:

Table

Measure

Lean-Eval result in the paper's evaluation

Lean-Eval problems solved with HOA251Lean-Eval result in the paper's evaluation
Methodology

Values come from the paper's stated benchmark experiments; they are not field measurements or general performance guarantees.

Updated:

Structured coverage

Subjects and places

Subject proposals

Geographic references

Only source-provided geographic references are shown. Opening an external map is optional. Optional story maps use OpenStreetMap. Opening a map sends your IP address and this site’s address to the map provider; location data remains readable as text without loading the map.

  • China · Place role: Coverage · Place type: Country or area · Location precision: Regional
    Evidence
    • Explicit preserved editorial country/region; no city or national centroid inferredarxiv.org ↗
Taxonomy version: IPTC Media Topics 2026-Q2

Continue reading

Technology

Rollout-Time Intervention tests agent training on new tasks

The framework adds multi-agent correction to training rollouts and filters for distribution shift.

arXiv preprint by Alibaba and Hong Kong University of Science and Technology researchers1 min read
Data included
Technology

Researchers alternate agent sessions on machine-learning tasks

Humanize MLE Agents alternates two agents, resetting session context while preserving a shared workspace.

arXiv preprint by Tsinghua University, Zhejiang University and collaborating researchers1 min read
Data included
Technology

ProgramGym builds 1,648 verifiable training environments for coding agents

The October 9 arXiv preprint describes environments for reproducing programs from scratch and a staged agent workflow.

arXiv preprint by researchers affiliated with Xiamen University, Washington University in St. Louis, the Chinese Academy of Sciences and Unisound1 min read
Data included