OhMyUnicorn lab · 10 November 2026

Claude Fable 5.2 directs Blender to build a warship, next to GPT-6 Astra and Gemini 4 Argon

On 14 October we gave GPT-6 Astra and Gemini 4 Argon one picture of a warship and the same short prompt: write Blender Python, render, look at your own render, fix it, repeat until it matches. Today the newest Claude model, Fable 5.2, gets the exact same picture and prompt at three effort levels. Time, tokens and API price are logged for every run, and every ship is a 3D model you can spin.

The three runs

All three runs are done. Results below.

Effort: lowwaiting

Effort: mediumwaiting

Effort: highwaiting

What we saw

Fable 5.2 finished all three runs on its own, with no stuck requests: 7.3 minutes for $2.14 at low, 12.3 minutes for $2.80 at medium, 32.6 minutes for $12.32 at high.
Low and medium cost about what Astra's did ($2.70 and $2.74). Low was the fastest run on this page that finished on its own (Astra 10.9 minutes, Gemini 4 in Antigravity 24). At medium, Astra was a minute quicker (11.4 against 12.3).
High is where the bill jumps: $12.32, about 2.3 times Astra's $5.24 and more than twice Gemini 4's $5.2 to $5.4. Fable wrote 175k output tokens (Astra: 32k) over 37 model calls, and re-read 5.9 million cached input tokens along the way. Most of that price is output and cache writes.
The ships are light and well split: 21k to 35k triangles in 23 to 89 named parts. Astra's ships have 52k to 156k triangles; Gemini 4 merges its ship into 8 to 17 parts.
Which ship looks best is your call. All three Fable ships show the whole hull with the helicopter deck, mast, stack, deck gun and missile cells; spin them against the others below. One flag from our automatic framing check: the ship fills 98 to 100% of the picture width in all three Fable renders, which it counts as touching the frame edge (Astra's high render passed).

Results

Effort	Model	Time	API price	Input tokens	Output tokens	Triangles	Parts	What happened
low	Claude Fable 5.2	7.3 min	$2.14	399,605	29,144	23,736	89	Finished on its own
low	GPT-6 Astra	10.9 min	$2.70	1,579,727	13,593	51,666	1129	Finished on its own
low	Gemini 4 Argon · Antigravity CLI	24 min	$2.80	3,553,754	98,567	60,992	8	Finished on its own
low	Gemini 4 Argon · Claude Code	6.4 min	$1.48	1,336,388	77,662	13,476	17	Model ready at 6.4 min, then a request hung; stopped at 50 min
medium	Claude Fable 5.2	12.3 min	$2.80	465,565	40,700	21,145	23	Finished on its own
medium	GPT-6 Astra	11.4 min	$2.74	1,463,536	14,592	75,872	1512	Finished on its own
medium	Gemini 4 Argon · Antigravity CLI	38 min	$3.05	3,777,691	149,065	23,266	12	Finished on its own
medium	Gemini 4 Argon · Claude Code	no model	$1.42	794,081	86,689	none		A request hung before it saved a model; stopped at 50 min
high	Claude Fable 5.2	32.6 min	$12.32	6,139,933	175,110	35,222	34	Finished on its own
high	GPT-6 Astra	18.5 min	$5.24	2,824,693	31,585	155,834	156	Finished on its own
high	Gemini 4 Argon · Antigravity CLI	40 min	$5.36	5,059,531	267,500	31,545	11	Finished on its own
high	Gemini 4 Argon · Claude Code	32.8 min	$5.21	3,600,859	349,371	61,572	12	Finished on its own

API price = tokens times list price per million tokens. Claude Fable 5.2: $8 input, $0.20 cached input, $10 or $16 to write the cache (5 minutes or 1 hour; Claude Code uses 1 hour), $40 output; we ran it on a Claude subscription, so for us this is an equivalent, not a bill. GPT-6 Astra: $10 input, $1 cached input, $50 output (ChatGPT subscription, also an equivalent). Gemini 4 Argon: $2 input, $0.20 cached input, $10 output. Input tokens include cached ones. The Astra and Gemini 4 rows are the 14 October runs, not rerun.

The picture and the prompt

The reference picture: a grey guided-missile cruiser at sea, seen from the front three-quarter — Made with gpt-image-2 from this description: A modern naval guided-missile cruiser (fictional class, no real hull numbers, no flags, no text), clean 3/4 front-side view from slightly above, full ship visible bow to stern, on calm dark-grey sea. Haze-grey hull with a sharp raked bow, a tall angular superstructure with flat phased-array radar panels, one mast with antennas, a stack, one forward deck gun, vertical launch cells fore and aft, a helicopter deck with hangar at the stern. Realistic, photographic lighting.

All models got this picture and this prompt, word for word (machine names replaced):

Recreate the warship in the reference image as a 3D model in Blender.
The reference is reference.jpeg in this directory: look at it with your image tool first.
Blender 5.2 is headless on another machine (it has an NVIDIA GPU; Cycles can use OPTIX). Write bpy scripts here, run them with:
  ssh <gpu-box> blender -b --python-exit-code 1 --python <remote-dir>/<script>.py
(scp your script to <gpu-box>:<remote-dir>/ first, scp renders back here, then LOOK at them with your image tool.)
Iterate until the render matches the reference: hull shape and bow, superstructure blocks and radar panels, mast, stack, deck gun, missile cells, flight deck and hangar, haze-grey paint.
Real-world scale in metres, separate named objects for the main parts.
Deliver in this directory: final.png (1920x1080 Cycles, similar 3/4 view) and cruiser.glb. Stop when you judge it matches.

Side by side, effort by effort

Left to right: Claude Fable 5.2, GPT-6 Astra, Gemini 4 Argon in Google's Antigravity CLI. Drag to turn a ship, scroll or pinch to zoom, right-drag to pan. Full screen shows the most detail. Under each viewer is the model's own Cycles render.

Effort: low

same picture, same prompt

Claude Fable 5.2

Finished on its own7.3 min$2.1429k output tokens23,736 triangles · 89 parts

GPT-6 Astra

Finished on its own10.9 min$2.7014k output tokens51,666 triangles · 1129 parts

Gemini 4 Argon · Antigravity CLI

Finished on its own24 min$2.8099k output tokens60,992 triangles · 8 partsIts 3D file was saved without paint colours, so the viewer shows it white

Effort: medium

same picture, same prompt

Claude Fable 5.2

Finished on its own12.3 min$2.8041k output tokens21,145 triangles · 23 parts

GPT-6 Astra

Finished on its own11.4 min$2.7415k output tokens75,872 triangles · 1512 parts

Gemini 4 Argon · Antigravity CLI

Finished on its own38 min$3.05149k output tokens23,266 triangles · 12 parts

Effort: high

same picture, same prompt

Claude Fable 5.2

Finished on its own32.6 min$12.32175k output tokens35,222 triangles · 34 parts

GPT-6 Astra

Finished on its own18.5 min$5.2432k output tokens155,834 triangles · 156 parts

Gemini 4 Argon · Antigravity CLI

Finished on its own40 min$5.36268k output tokens31,545 triangles · 11 parts

Gemini 4 in Claude Code (first round)

the harness Fable 5.2 ran in; low and medium got stuck

Gemini 4 Argon · Claude Code · low

Model ready at 6.4 min, then a request hung; stopped at 50 min6.4 min$1.4878k output tokens13,476 triangles · 17 parts

Gemini 4 Argon · Claude Code · medium

A request hung before it saved a model; stopped at 50 minno model$1.4287k output tokens

No 3D model: the run stopped before it saved one.

Its last draft render (it never saved a final one) by Gemini 4 Argon · Claude Code · medium — Its last draft render (it never saved a final one)

Gemini 4 Argon · Claude Code · high

Finished on its own32.8 min$5.21349k output tokens61,572 triangles · 12 parts

How we ran it

Claude Fable 5.2 ran in Claude Code (model claude-fable-5-2) at low, medium and high effort, with clean settings: none of our private instructions, memory or add-on tools, only the prompt and the picture. It had shell access, could run Blender 5.2 headless on a GPU server, and could open its own renders as images.
The three runs started within a minute of each other on the same GPU server as the October runs, with a 50-minute limit each. Nobody stepped in during a run.
The GPT-6 Astra and Gemini 4 Argon results are the 14 October runs (Astra in the Codex CLI, Gemini 4 in Claude Code and in Google's Antigravity CLI). They were not rerun today, so models and harnesses may have moved a little since.
Effort is each model's own reasoning setting (low, medium, high).
One run per setting. Expect run-to-run variation; treat this as a field note, not a benchmark. Judging the ships is your call: we only measured time, tokens and price.