ORO Live Demo

Model performance · week over week

success_rate field_matching format_score published

success_rate · latest

—

field_matching · latest

—

since first run

—

Runs

Week	Stages	success	field_match	Gate

Data volume · DPO pairs (balanced)

Published model

run—

success_rate—

eval set—

checkpoint—

gatepublished

Try the model

now in Live Demo tab

Head over to the Live Demo tab above — pick a curated ShoppingBench problem (or type your own), and watch our SFT-tuned Qwen3-4B race Fable 5, Kimi K3, and GPT-5.6 Sol side-by-side in real time.

Reads run manifests from S3 · AWS-hosted · access-gated. No agent code or secrets shown.

Pick a shopping task from the ShoppingBench eval set, or type your own query. We run ORO's fine-tuned Qwen3-4B checkpoint against Claude Fable 5, Kimi K3, and GPT-5.6 Sol on the same task at the same time. Each column streams the model's reasoning, tool calls, and final recommendation as they arrive. When the task has known-correct answers, we score every run with the ShoppingBench grader.

Loading budget…

Bring your own OpenRouter key Your access code doesn't include a bundled OpenRouter budget. Paste your own key below to run the demo. It's kept in this tab's memory only — never sent to our servers for storage and never written to logs.

Problem

Free-form query (overrides problem)

—