LiveMACE tests five frontier agents over 30 days of live markets

What the research found

LiveMACEBench evaluates five frontier LLM agents over 30 days of live financial markets and finds a gap between realized returns and capability-specific diagnostics from decision traces.

Claims and how well the evidence supports them

1. Five frontier LLMs evaluated over 30 days of live financial market trading 98%
2. Outcome-capability gap found between realized returns and capability measures 99%
3. Evaluation covers Tool Use, Persistent Memory, Rule Following and Multi-Agent Collaboration configurations 99%
4. Benchmark is named LiveMACEBench 99%

Evidence

• huggingface.co https://huggingface.co/papers/2610.09872

Actionability rated low.