LetTokenBurn
Sign in
Verified86%Harbor: pendingAgentic / Tool UseCoding

FDE-Gym

FDE-Gym: The Last Mile of AI Delivery

FDE-Gym: The Last Mile of AI Delivery FDE-Gym tests whether an agent can do the work of a forward-deployed engineer: enter a customer's running environment, find out what matters there, make the decisions the job needs, and ship a system that holds up in production.

PaperDataset

How to run

ltb · fde-gym
$ npm i -g lettokenburn-cli
$ ltb bench info fde-gym

External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.

Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs

Scores

9 sourced entries
HarnessSourceConfidenceLink
Claude Opus 5.50.6N/AIndependentmediumOct 15, 2026arxiv.org
Claude Sonnet 5.50.5N/AIndependentmediumOct 15, 2026arxiv.org
GPT-6 Astra0.4N/AIndependentmediumOct 15, 2026arxiv.org
GPT-6.1 Sol0.4N/AIndependentmediumOct 15, 2026arxiv.org
GLM-5.3 Flash0.2N/AIndependentmediumOct 15, 2026arxiv.org
DeepSeek V4.1 Flash0.2N/AIndependentmediumOct 15, 2026arxiv.org
Gemini 3.8 Flash0.2N/AIndependentmediumOct 15, 2026arxiv.org
Qwen3.8 Flash0.1N/AIndependentmediumOct 15, 2026arxiv.org
GPT-5.50.0N/AIndependentmediumOct 15, 2026arxiv.org

Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.

Reproduced by the community

All runs
Loading community runs…

About

FDE-Gym: The Last Mile of AI Delivery

FDE-Gym tests whether an agent can do the work of a forward-deployed engineer: enter a customer's running environment, find out what matters there, make the decisions the job needs, and ship a system that holds up in production. Paper: arXiv:2610.22350 · Leaderboard: https://fde-gym.com/paper This repository holds the original task of each of the 20 scenarios.

What is here

cases/<scenario>/ engagement.json title, what the…