Feed
researchAI Agent Research1d ago

Researchers introduce GRACE, a multi-agent pipeline for tabular data analysis

Facts

GRACE (Generative, Ranked, Adaptive Code Ensemble) is a new multi-agent pipeline designed to improve the accuracy of LLM-based analysis of tabular dataframes. By using a parallel best-of-N approach with a tournament-style selection and adaptive retry mechanism, the system demonstrated significant performance gains on the InfiAgent-DABench and DataBench benchmarks.

  • ▸GRACE uses a pairwise tournament to evaluate and select the best candidate output from multiple parallel agents.
  • ▸The system supports an abstention mechanism that allows it to reject outputs when no defensible candidate is found.
  • ▸On InfiAgent-DABench using Qwen3.6-35B-A3B, GRACE increased answer match rates from 39% to 71%.
  • ▸Performance improvements were validated across both the Qwen3.6-35B-A3B and Qwen3.5-9B models.
  • ▸The research identified a tradeoff between parallel agent count and retry budget, allowing for cost-optimized deployment.
Occurred
2026-11-16 12:16 UTC
First reported
2026-11-16 12:16 UTC
Last updated
2026-11-17 08:16 UTC
Event type
research
Region
global

Entities

Categories

Source timeline · 1

  • arXivFirst-party2026-11-16 12:16
Community vote

Sign in to vote on importance and sentiment.

Platform Signals

Impact score
66
AI sentiment
Positive
Event significance30%1
Entity importance20%1
Source credibility20%1
Novelty15%1
Corroboration15%1
Confidence1
Novelty1
Corroboration1
Community
ImportanceNo votes
Strategic sentiment
Positive 0%Neutral 0%Negative 0%

Platform signals are algorithmic and community-derived indicators. They are not investment recommendations and do not predict valuation or performance.