Did Codex Reset
GitHub

Self-organizing team of three models averages 66.7% in Stanford and Together AI paper

DAIR.AI

DAIR.AI reports that a paper from Stanford and Together AI tested o3-mini, Claude Sonnet 4, and DeepSeek-V3 as a self-organizing team. Across five math and physics benchmarks, the team averaged 66.7%. The strongest member alone scored 48.8%, and a perfect router choosing among the members' independent answers scored 59.0%.

On AIME 2026, the team reached 71.2%, 13.4 points above that router. One member reviews the team's earlier exchanges and rewrites the teamwork strategy, covering roles, the order of discussion phases, who participates, and how partial answers are combined. Those strategies were learned from 15 AIME 2024 problems and then applied unchanged to held-out AIME 2024 problems and four new benchmarks.

#ai-research#multi-agent#benchmarks#stanford#together-ai

Only 35.8% of the cleanest public terminal-agent RL collection passed a Salesforce AI Research audit

Salesforce AI Research put that collection at 35.8% clean and two others at 10.1% and 3.3%, after finding reward errors in both directions. The authors say RIVER, which filters defective environments, increases RL gains by 106% on Terminal-Bench-Lite and 30% on Terminal-Bench v2.1 while using fewer than 30% of TMax's environments.