Sakhanda Wire
NVDA $154.51 -1.04% MSFT $506.49 -2.22% GOOGL $281.27 -0.86% META $621.51 -0.78% AMZN $256.90 -0.33%
← Back to the news

NVIDIA’s SoL-Pi cuts coding-agent token traffic by up to 49%

Researchers from NVIDIA, NTU and MIT have introduced SoL-Pi, an MIT-licensed extension for the open-source Pi coding agent. Its four mechanisms were discovered by an AI through automated research loops that tested harness changes across 535 environments, including GitHub issue-pull request tasks and synthetic tasks. The extension runs with an unmodified Pi release and was tested with Pi 0.85.1 and Node.js 22.19 or newer.

On the 51-task EdgeBench evaluation, the full system reduced token traffic by 49.0% with GPT-5.6 Sol and by 44.7% with Opus 5, compared with Pi. API costs fell by 33.2% and 33.5%, respectively, while average scores remained at 93.7% and 94.3% of Pi’s results. The mechanisms combine tool actions, compact context when beneficial, summarize large observations while retaining retrieval, and reduce build or test logs through verified receipts.

The search examined 152 directions and more than 3,000 runs. On Terminal-Bench 4, SoL-Pi solved 15 of 63 tasks, compared with 18 for Pi, while reducing cost by 26.3%. The researchers described transfer to Opus 5 as preliminary because the search used GPT-5.6 Sol trajectories.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news