A GPT-5.6 creative-coding experiment

The ThinkRun story,
40 ways.

ThinkRun was built across 323 days. We wanted to see how well GPT-5.6 could turn that same development record into a compelling launch story—and whether more reasoning effort produced better results. So GPT-5.6 Sol and Terra each received the same repository context and creative brief 20 times. These are all 40 outcomes, including the runs that produced no page.

01 Each poster is one independent run 02 Open it to inspect the real webpage 03 “No ship” means the run produced no page

The result

Here’s what we learned.

Higher reasoning effort did not make this task more reliable. The clearest way to judge creative quality is to open the same brief’s outputs and decide which story you would ship.

19/20Medium-effort runs shipped

Only one medium-effort run ended without a webpage.

14/20High-effort runs shipped

Six high-effort runs produced no reviewable page.

+60%More normalized usage at high effort

$25.13 at high effort versus $15.75 at medium.

Source context

What the model could inspect before it designed.

Every run received the same history-free snapshot of the relevant ThinkRun working repository. The model chose which material to read and how to turn it into a story.

Design system + tokensThinkRun colors, typography, interface patterns, and visual references.

Website + frontendCurrent product pages, UI assets, product language, and frontend structure.

Docs + product blogFeature explanations, positioning, technical documentation, and published narratives.

Development memory wikiVerified milestones, dates, incidents, decisions, corrections, and lessons from the build.

Selected project codeRepository structure and lightweight source/configuration relevant to the product story.

One identical creative briefBuild one self-contained ThinkRun launch-story page; invent no statistics.

Anti-copy controls: prior generated story pages, git branches/history, recordings, dependency folders, agent transcripts, and heavy service/build directories were removed. Every statistic in a page was required to come from the supplied docs or memory.

40verified cells

33pages shipped

7valid no-ships

$40.88normalized task usage

40 results · $40.88

GPT-5.6 sol

high effort

7/10 shipped • $12.83 usage

Preview of ThinkRun — The browser remembers

$1.17 usage • 53 min • 46 KB

GPT-5.6 sol • high • Run 1ThinkRun — The browser remembers$1.17 · 53 min · Shipped

Valid no-ship

$0.42 usage • 1 min • 0 KB

Terminal outcome

GPT-5.6 sol • high • Run 2No webpage artifact$0.42 · 1 min · Valid no-ship

Valid no-ship

$0.99 usage • 2 min • 0 KB

Terminal outcome

GPT-5.6 sol • high • Run 3No webpage artifact$0.99 · 2 min · Valid no-ship

Preview of ThinkRun — The browser learned to remember

$1.33 usage • 13 min • 34 KB

GPT-5.6 sol • high • Run 4ThinkRun — The browser learned to remember$1.33 · 13 min · Shipped

Preview of ThinkRun — The browser became an instrument

$1.35 usage • 53 min • 34 KB

GPT-5.6 sol • high • Run 5ThinkRun — The browser became an instrument$1.35 · 53 min · Shipped

Preview of ThinkRun — The browser learned to remember

$1.40 usage • 52 min • 46 KB

GPT-5.6 sol • high • Run 6ThinkRun — The browser learned to remember$1.40 · 52 min · Shipped

Preview of ThinkRun — Browser Archaeology

$0.66 usage • 7 min • 25 KB

GPT-5.6 sol • high • Run 7ThinkRun — Browser Archaeology$0.66 · 7 min · Shipped

Valid no-ship

$0.80 usage • 3 min • 0 KB

Terminal outcome

GPT-5.6 sol • high • Run 8No webpage artifact$0.80 · 3 min · Valid no-ship

Preview of ThinkRun — The browser finally remembers

$1.84 usage • 30 min • 55 KB

GPT-5.6 sol • high • Run 9ThinkRun — The browser finally remembers$1.84 · 30 min · Shipped

Preview of ThinkRun — 323 days to teach software how to see

$2.87 usage • 57 min • 45 KB

GPT-5.6 sol • high • Run 10ThinkRun — 323 days to teach software how to see$2.87 · 57 min · Shipped

GPT-5.6 sol

medium effort

9/10 shipped • $8.38 usage

Valid no-ship

$0.53 usage • 2 min • 0 KB

Terminal outcome

GPT-5.6 sol • medium • Run 1No webpage artifact$0.53 · 2 min · Valid no-ship

Preview of ThinkRun — The browser became a memory

$1.16 usage • 12 min • 28 KB

GPT-5.6 sol • medium • Run 2ThinkRun — The browser became a memory$1.16 · 12 min · Shipped

Preview of ThinkRun — The browser finally learned to explain itself

$0.52 usage • 51 min • 21 KB

GPT-5.6 sol • medium • Run 3ThinkRun — The browser finally learned to explain itself$0.52 · 51 min · Shipped

Preview of ThinkRun — The Run That Built Itself

$0.75 usage • 5 min • 29 KB

GPT-5.6 sol • medium • Run 4ThinkRun — The Run That Built Itself$0.75 · 5 min · Shipped

Preview of ThinkRun — We gave the browser a memory

$0.95 usage • 53 min • 20 KB

GPT-5.6 sol • medium • Run 5ThinkRun — We gave the browser a memory$0.95 · 53 min · Shipped

Preview of ThinkRun — The browser remembers

$1.05 usage • 26 min • 18 KB

GPT-5.6 sol • medium • Run 6ThinkRun — The browser remembers$1.05 · 26 min · Shipped

Preview of ThinkRun — The browser remembers

$0.65 usage • 24 min • 21 KB

GPT-5.6 sol • medium • Run 7ThinkRun — The browser remembers$0.65 · 24 min · Shipped

Preview of ThinkRun — The browser was the missing context

$1.19 usage • 60 min • 25 KB

GPT-5.6 sol • medium • Run 8ThinkRun — The browser was the missing context$1.19 · 60 min · Shipped

Preview of ThinkRun — The browser learned to remember

$0.46 usage • 4 min • 23 KB

GPT-5.6 sol • medium • Run 9ThinkRun — The browser learned to remember$0.46 · 4 min · Shipped

Preview of ThinkRun — The browser learned to speak agent

$1.13 usage • 21 min • 26 KB

GPT-5.6 sol • medium • Run 10ThinkRun — The browser learned to speak agent$1.13 · 21 min · Shipped

GPT-5.6 terra

high effort

7/10 shipped • $12.31 usage

Preview of ThinkRun — The evidence has a memory

$2.99 usage • 35 min • 36 KB

GPT-5.6 terra • high • Run 1ThinkRun — The evidence has a memory$2.99 · 35 min · Shipped

Preview of ThinkRun — The browser remembers

$1.25 usage • 24 min • 25 KB

GPT-5.6 terra • high • Run 2ThinkRun — The browser remembers$1.25 · 24 min · Shipped

Valid no-ship

$0.88 usage • 3 min • 0 KB

Terminal outcome

GPT-5.6 terra • high • Run 3No webpage artifact$0.88 · 3 min · Valid no-ship

Preview of ThinkRun — The witness machine

$0.85 usage • 21 min • 23 KB

GPT-5.6 terra • high • Run 4ThinkRun — The witness machine$0.85 · 21 min · Shipped

Preview of ThinkRun — The build log

$1.60 usage • 10 min • 28 KB

GPT-5.6 terra • high • Run 5ThinkRun — The build log$1.60 · 10 min · Shipped

Valid no-ship

$0.82 usage • 3 min • 0 KB

Terminal outcome

GPT-5.6 terra • high • Run 6No webpage artifact$0.82 · 3 min · Valid no-ship

Preview of ThinkRun — The browser became a story

$1.90 usage • 10 min • 28 KB

GPT-5.6 terra • high • Run 7ThinkRun — The browser became a story$1.90 · 10 min · Shipped

Valid no-ship

$0.48 usage • 3 min • 0 KB

Terminal outcome

GPT-5.6 terra • high • Run 8No webpage artifact$0.48 · 3 min · Valid no-ship

Preview of ThinkRun — 323 days to teach software how to see

$0.77 usage • 51 min • 24 KB

GPT-5.6 terra • high • Run 9ThinkRun — 323 days to teach software how to see$0.77 · 51 min · Shipped

Preview of ThinkRun — The Flight Recorder

$0.77 usage • 51 min • 34 KB

GPT-5.6 terra • high • Run 10ThinkRun — The Flight Recorder$0.77 · 51 min · Shipped

GPT-5.6 terra

medium effort

10/10 shipped • $7.37 usage

Preview of ThinkRun — The browser learned to remember

$0.83 usage • 5 min • 27 KB

GPT-5.6 terra • medium • Run 1ThinkRun — The browser learned to remember$0.83 · 5 min · Shipped

Preview of ThinkRun — The browser has a memory

$0.90 usage • 4 min • 18 KB

GPT-5.6 terra • medium • Run 2ThinkRun — The browser has a memory$0.90 · 4 min · Shipped

Preview of ThinkRun — The Browser Learned to Tell You What Happened

$0.58 usage • 13 min • 24 KB

GPT-5.6 terra • medium • Run 3ThinkRun — The Browser Learned to Tell You What Happened$0.58 · 13 min · Shipped

Preview of ThinkRun — a browser learns to tell the truth

$0.87 usage • 34 min • 19 KB

GPT-5.6 terra • medium • Run 4ThinkRun — a browser learns to tell the truth$0.87 · 34 min · Shipped

Preview of ThinkRun — The browser was always the interface

$0.92 usage • 4 min • 21 KB

GPT-5.6 terra • medium • Run 5ThinkRun — The browser was always the interface$0.92 · 4 min · Shipped

Preview of ThinkRun — The browser needed a witness

$0.68 usage • 40 min • 19 KB

GPT-5.6 terra • medium • Run 6ThinkRun — The browser needed a witness$0.68 · 40 min · Shipped

Preview of ThinkRun — The build log, made visible

$0.56 usage • 12 min • 19 KB

GPT-5.6 terra • medium • Run 7ThinkRun — The build log, made visible$0.56 · 12 min · Shipped

Preview of ThinkRun — The browser learns to remember

$0.79 usage • 23 min • 27 KB

GPT-5.6 terra • medium • Run 8ThinkRun — The browser learns to remember$0.79 · 23 min · Shipped

Preview of ThinkRun — A browser learned to listen

$0.72 usage • 21 min • 22 KB

GPT-5.6 terra • medium • Run 9ThinkRun — A browser learned to listen$0.72 · 21 min · Shipped

Preview of ThinkRun — The browser became a teammate

$0.50 usage • 4 min • 22 KB

GPT-5.6 terra • medium • Run 10ThinkRun — The browser became a teammate$0.50 · 4 min · Shipped

No matching runsTry a broader filter combination.