oneshotlm
Compare all models on this prompt →

Force-directed graph · gpt-4.1-nano

Using the pre-provided d3.min.js (global d3), render an interactive force-directed graph of about 40 nodes in several clusters with draggable nodes, hover highlighting of neighbors, and zoom/pan. Define the sample nodes and links inline.

⛨ sandboxed Open in new tab ↗
transcript · 18 messages
user

Using the pre-provided d3.min.js (global d3), render an interactive force-directed graph of about 40 nodes in several clusters with draggable nodes, hover highlighting of neighbors, and zoom/pan. Define the sample nodes and links inline.

assistant
read {"path":"index.html"}
write index.html (0 bytes)
toolResult

ENOENT: no such file or directory, access '/tmp/ws/index.html'

toolResult

Successfully wrote 0 bytes to index.html

assistant
read {"path":"index.html"}
read {"path":"d3.min.js"}
toolResult

toolResult

// https://d3js.org v7.9.0 Copyright 2010-2023 Mike Bostock [Showing lines 1-1 of 3 (50.0KB limit). Use offset=2 to continue.]

assistant
read {"path":"index.html"}
read {"path":"d3.min.js"}
toolResult

toolResult

// https://d3js.org v7.9.0 Copyright 2010-2023 Mike Bostock [Showing lines 1-1 of 3 (50.0KB limit). Use offset=2 to continue.]

assistant
read {"path":"index.html"}
read {"path":"d3.min.js"}
toolResult

toolResult

// https://d3js.org v7.9.0 Copyright 2010-2023 Mike Bostock [Showing lines 1-1 of 3 (50.0KB limit). Use offset=2 to continue.]

assistant
read {"path":"index.html"}
toolResult

assistant
write index.html (2929 bytes)
toolResult

Successfully wrote 2929 bytes to index.html

assistant

I have created a complete self-contained `index.html` file that renders an interactive force-directed graph with about 40 nodes, clustered, draggable, with hover highlighting and zoom/pan functionality. You can open this file in a browser to see the visualization.

TAKE IT TO CODE

A single self-contained HTML file — copy it and open in any browser, or reuse the prompt with your own pi-agent run.

RUN

modelgpt-4.1-nano
vendorOpenAI
harnesspi-agent
generated2026-08-01

EVALUATION

3/5 partial

Force-directed clustered graph renders correctly; drag and wheel slightly move/pan nodes, but graph is tiny and off-center in viewport.

Vision grade · anthropic/claude-sonnet-5

MEASUREMENTS

Raw readings from the capture, not a verdict — the grade above is the only judgement. Percentages are how much of the screen changed; the measure under-reports sparse motion on a dark background.

idle motion0.0%
on click0.0%
on drag5.8%
on wheel2.9%
on enter + space0.0%
on arrow keys0.0%
on w/a/s/d0.0%
frame spread10.5 / 255
console errors0
js errors none

METRICS

runtime11.9s
tokens in11k
tokens out1.3k
cached6.5k
cost / run$0.0012