Xiuyu Li

@sheriyuo

Researcher @StepFun_ai | Working on long-horizon tasks | Opinions are my own

following
followers

RSIGym launches as an open-source research environment for AI agents

Evolvent AI says Opus 5 led its RSI-Index study of six research agents across five benchmarks, scoring 0.4809.

4 Sources, 12d ago, updated 12d ago

StepFun introduces Step 5 Preview, with open weights planned for October 15, 2026

StepFun says its new flagship model has 600 billion total parameters, 27 billion active parameters, a 1M context window and vision support. It emphasizes software engineering and professional knowledge work, claiming particular strength in finance.

猫C

7 Sources, 30d ago, updated 30d ago

AGI Frontier event in Shanghai

An attendee reported being at AGI Frontier in Shanghai, naming @agihouse_org as the host.

2 Sources, 31d ago, first seen 31d ago

Researcher Questions Post-Training Benchmark Reliability

Xiuyu Li notes how evaluations can be manipulated through settings and harness choices.

2 Sources, 46d ago, first seen 46d ago

Featured in

RSIGym launches as an open-source research environment for AI agents

StepFun introduces Step 5 Preview, with open weights planned for October 15, 2026

AGI Frontier event in Shanghai

Researcher Questions Post-Training Benchmark Reliability

Featured in

RSIGym launches as an open-source research environment for AI agents

StepFun introduces Step 5 Preview, with open weights planned for October 15, 2026

AGI Frontier event in Shanghai

Researcher Questions Post-Training Benchmark Reliability