NVIDIA open-sources SoL-Pi: auto-research loops cut agent token cost by up to 54%
udmrzn · x · 2026-09-11
NVIDIA's NVLabs open-sourced SoL-Pi (MIT license), a Pi-based agent-harness efficiency layer discovered via scaled auto-research loops. It builds 535 executable training environments (495 from real GitHub Issue-PRs, 40 synthetic with verifiers) and lets an auto-research loop propose, implement, and validate harness improvements — only 1/40 of ideas survive.
Four winning optimizations: Action Fusion (merge test/run into one tool call), Online Context Compact (compact at subtask completion points), ObservationPack (index oversized tool results instead of replaying them), and Evidence-Preserving Reducer (cheap agent compresses long logs while cited evidence is verified item by item).
Results: vs original Pi, 45–49% fewer tokens and 1/3 lower cost at 94% of average task score; vs native harnesses, 35–64% fewer tokens and 50–54% lower API-priced cost. On GPT-5.6 Sol, SoL-Pi even beats the native Codex harness on EdgeBench. The author notes this converges with DeepSeek's harness direction.
Related event: NVIDIA Open-Sources SoL-Pi, Cutting Pi Coding Agent Token Costs(2 posts)→
More from coding & agent
- Indie game dev: AI handles hundreds of UI variables so he can focus on the craft — round · 2026-09-11
- CursorBench 4.0 rolls out with harder, longer-horizon coding tasks, scores drop — StringChaos · 2026-09-11
- How do you route long-running agents across models after a cost shift? — Katleen_Cole · 2026-09-11
- Coding Agents talk at KCDC: start simple, scale smart, says developer Dan Vega — therealdanvega · 2026-09-11
- Run OpenAI Agents API sessions in Daytona sandboxes with zero inbound ports — mattturck · 2026-09-11
- 95% local AI + 5% GPT-6: a full basketball tracking pipeline breakdown — TheMoonMidas · 2026-09-11