Skip to Main Content

GTerm - Small Model Big Bench

Google I/O Hackathon

How high can Gemini 3.5 Flash climb on Terminal-Bench? GTerm is an attempt to answer this question. We believe we can beat SOTA! Early results are promising. How it works: GTerm is running a Meta-Harness-inspired self-improving loop on Flash. The original example in the Meta-Harness paper reached 79.2% with Claude Code + Opus 4.7 improving over ten continuous days of harness evolution. We’re on Day 1 with Flash and started at 37% and are climbing. Smaller model. New loop. Big bench.

Team

dominicMolt

Placement

Participant

More from Google I/O Hackathon

Billie Gene

Billie Gene is a rapid vaccine-design research acceleration platform built to reduce the early bottleneck in responding to emerging viral outbreaks such as…

Billie Gene project preview

Sky Guardian

Our technology has reached the point where society needs to stop putting human lives at risk and start leveraging the tools around us. Picture this: a…

Sky Guardian project preview

AgentGym

An RL environment to help train your managed agents in your voice.

us06web.zoom.us/…

Autonomous Infrastructure Auditor

1 in 4 Americans lives with a disability, yet most cities have no systematic way to audit pedestrian infrastructure for ADA compliance at scale. Manual…

Autonomous Infrastructure Auditor project preview

Gemini Lifeguard Rescue

Gemini Lifeguard Rescue is an AI-powered water safety system that helps lifeguards detect and respond to emergencies faster. The demo focuses on beach rescue…

Gemini Lifeguard Rescue project preview

Autonomous Player Fleet

Autonomous Android Simulator Player Fleet & Telemetry Ingest It is a premium, real-time platform designed to run autonomous gameplay testing agents on a…

Autonomous Player Fleet project preview