Inkling 2 Mini, same picture

One picture, four 3D worlds. Thinking Machines' Inkling 2 Mini in OpenCode at four effort levels.

Thinking Machines' Inkling 2 Mini, running locally on one GPU, built the island cabin picture as a Three.js world on low, medium, high and extra high effort, from the same prompt and the same copy of Three.js as the other runs. One run each, with no human edits. High and extra high stopped at the 60-minute limit, and extra high's world is blank.

This is not a CmdBench benchmark run: no checks, scores or verdicts.

AI-generated picture of a miniature island: a wooden cabin with a stone chimney, a jetty and rowboat, a yellow crane over a half-built timber frame, lumber stacks, rocks and pine trees, in warm late-afternoon light.
Reference picture, AI-generated. Every run got this picture and the same prompt.

The four worlds

One run on each effort level. Each world is served exactly as the model left it, with no fixes, and opens full screen in this tab.

  • Frame from the high effort build, captured from the build itself. Opens the world.

    High effort

    Stopped at the 60-minute limit. The world is served as it was left.

    Built by
    Inkling 2 Mini, high effort
    Time
    60m 00s
    Output tokens
    At least 137,006
    Cost
    None, local model

    Drag to orbit, scroll or pinch to zoom, and move the Sun and Wind sliders.

  • Frame from the extra high effort build, captured from the build itself. Opens the world.

    Extra high effort

    Stopped at the 60-minute limit. The world is blank: the page loads a file the model never wrote, so only the Sun and Wind sliders show.

    Built by
    Inkling 2 Mini, extra high effort
    Time
    60m 00s
    Output tokens
    At least 119,953
    Cost
    None, local model

Two differences from the Claude runs: OpenCode sends the model its own system prompt, and the model ran locally on one GPU, not through an API. The tools were the same file tools, and the picture went to the model unchanged.

No costs: the model ran locally on one GPU, with no API.

“At least”: the run ended during the model's last reply, and that reply has no token count.

How it was made

Picture
AI-generated. The exact PNG every run got
Prompt
The same for every run: read the prompt (TASK.md)
Agent
OpenCode 2.3.2
Effort levels
low, medium, high and extra high; one run each
Tools
Only read, write, edit, glob and grep: no shell, no browser, no web. The model never saw its world running.
Time limit
60 minutes per run
Three.js
0.187.0
Human edits
None. Each world is served exactly as the agent wrote it, with no fixes.
Run date