RSIAgent: Zero Training, Just Environment Exploration Beats GPT-6 on Agent Benchmarks
大模型之路 · wechat · 2026-09-24
RSIAgent (arXiv, Sep 14) is a training-free agent framework: it explores unfamiliar environments first — broad parallel exploration followed by targeted deep-dives, coordinated by a curriculum agent, an executor, and a verifier — then freezes an environment-specific memory for downstream tasks. The authors claim this lets Kimi-K3 and GLM-5.3 beat GPT-6 on OSWorld-v2 and Agent's Last Exam (self-run comparison, not third-party). The core thesis: agent capability gaps often lie in environment knowledge, not model weights, shifting the lever from compute to engineering. Best suited to stable interfaces with diverse tasks; frozen maps go stale under frequent UI releases and need periodic re-exploration.
Related event: RSIAgent: Training-Free Self-Improvement Beats GPT-6 with Open Models(3 posts)→
More from coding & agent
- Founder open-sources 'company brain', a one-click Slack AI employee harness — julianweisser · 2026-09-26
- Running 3 agents + 10 subagents in 16GB RAM with 5GB to spare — Teknium · 2026-09-26
- Dev swaps DeepSeek Flash for Opus 5.5 to fix endless bug loops — oran_ge · 2026-09-26
- FPS fighting Xenomorph-like aliens on a moon base, built with Opus 5.5 — EricBuess · 2026-09-26
- Agent orchestrator app adds ⌘+tab-style session switching like macOS — weswinder · 2026-09-26
- Luke Wroblewski: the best way to learn to use agents is watching how other humans use them — LukeW · 2026-09-26