NVIDIA Introduces SoL-Pi: Auto-Research Loops That Cut Coding Agent Token Traffic by Up to 49%
Coding agents no longer run for minutes—they run for hours. Every edit, test execution, and log read feeds back into the model's context, compounding cost. To address this, a research team from NVIDIA, NTU, and MIT has released SoL-Pi, a set of four efficiency mechanisms for the open-source Pi coding agent. The mechanisms were discovered by an AI running auto-research loops at the harness layer.
On the 51-task EdgeBench evaluation, SoL-Pi reduces recorded token traffic by 44.7% to 49.0% versus Pi and cuts API costs by roughly 33%—while maintaining scores close to Pi on both GPT-5.6 Sol and Opus 5.
Is it deployable? Yes. SoL-Pi ships on GitHub under NVlabs as an MIT-licensed extension that runs on an unmodified Pi release. It is tested with Pi 0.85.1 and Node.js 22.19 or newer.
Why Target the Harness
Most efficiency work reduces the cost per token through faster kernels, quantization, or cheaper models. SoL-Pi instead reduces how many tokens a task consumes. The harness is the layer that handles tool calls, context, observations, and delegation.
Tuning a harness by hand is slow, and its components are tightly coupled: a fix in one place can push cost into later steps. Meta-Harness and similar systems automate this work. However, a recent study found that evolved harnesses can overfit their search tasks and deliver only marginal gains on unseen ones.
How the Search Works
A research AI observes execution traces from a separate agent running base Pi. It then proposes harness changes and tests them. The search covered:
- 152 proposed directions across 6 families: context, progress, tools, delegation, prompt and policy, and improvement and evaluation
- 535 executable environments: 495 built from GitHub issue–pull request pairs and 40 synthetic tasks with executable verifiers
- More than 3,000 runs and 60,000+ agent-environment interactions
Each search is a disposable, isolated loop. It follows the autoresearch cycle, extended with a Ralph Wiggum-style approach to iterative refinement.
The Four Efficiency Mechanisms
SoL-Pi's four mechanisms target different points in the agent loop, from how context is assembled to how sub-tasks are delegated. Together they reduce redundant token generation without altering the underlying model or requiring a custom Pi fork.
Availability
SoL-Pi is open source under the MIT license. It runs on unmodified Pi 0.85.1 with Node.js 22.19 or newer, making it straightforward to adopt in existing coding agent workflows.
Paper: SoL-Pi on arXiv
Code: GitHub - NVlabs/SoL-Pi
via MarkTechPost