Today for AI
HOT RADAR
agentHEAT 11.8°

AI CURATED HOT STORY · 10/19/2026

StartLux Launches 27B Decision Model, Tops Leaderboard with Dynamic Reasoning

1 reports archived1 independent sourcesupdated 10/19/2026, 20:46:54
Synthesis & Latest Updates
1 Sources Cross-Validated

Chinese team StartLux released a 27B decision model that ranks first on the Hugging Face Decision Index 0.3.2, significantly outperforming Jev. The model introduces an AutoThink mechanism and RLCD² reinforcement learning method, enabling agents to dynamically adjust reasoning depth based on confidence levels, thereby improving decision reliability while controlling computational costs.

LATEST/Chinese team StartLux released a 27B decision model that ranks first on the Hugging Face Decision Index 0.3.2, significantly outperforming Jev. The model introduces an AutoThink mechanism and RLCD² reinforcement learning method, enabling agents to dynamically adjust reasoning depth based on confidence levels, thereby improving decision reliability while controlling computational costs.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. InfoQ 中文T2·86 pts
    • StartLux-Decision-27B ranked first across public benchmarks, private tests of similar capabilities, and new domain private tests with a composite score of 69.12.
    • The model demonstrates a ~27% relative lead over Jev's hosted version in Decision Index 0.3 evaluations, validating the advantage of specialized decision models.
    • Introduces dual paths: Instant (single-pass fast calculation) and AutoThink (confidence-based additional reasoning) to resolve the latency-precision trade-off in agent action selection.