Agima: An Imaginative Agent That Plans by Simulating Futures, 2%→40% on Hard Tasks
RulinShao · x · 2026-10-24
Researchers introduce Agima, an omni agent that plans by imagining futures. With SFT it directly generalizes to harder tasks, jumping from 2% to 40% and beating GPT-5.6 Sol (20%). With RL, new imagination strategies emerge—imagining a second plan only if the first fails, lifting scores from 32% to 100%.
Related event: Agima: Berkeley-led agent plans by imagining the future(2 posts)→
More from coding & agent
- Stop asking agents for text PR summaries: use diagrams, screen recordings, and specs — _AustinCalvert_ · 2026-10-24
- Dev experiments with autoencoders over functions; low-dim search may be key to local FunSearch — cephaloform · 2026-10-24
- Agent with SSH access successfully upgrades an ancient PHP 5.6 server end-to-end — DanielLockyer · 2026-10-24
- DeepMind researcher demos a playable 3D racing game built right on her desk — thursdai_pod · 2026-10-24
- Cut LLM costs: route requests, disable thinking, cache answers, escalate only on low confidence — colinmcnamara · 2026-10-24
- Evo Agents: Open-Source MCP Shared Context for Coding Agents Across Repos — binhbumpro · 2026-10-24