Claude Opus 5.5 vs GPT-6.1 Sol.

This review puts two flagship models through the same creative-coding job: write a Blender script that builds, lights, animates, and renders a 30-second 3D film. The contenders are Anthropic's Claude Opus 5.5 at Max effort and OpenAI's GPT-6.1 Sol at its Ultra code setting. We are not naming a winner. The goal is to show both results and let you decide which one fits your own work.

If you want the individual models' profiles, see our listing for Claude Opus 5 and our GPT-6.1 Sol launch coverage.

What we tested

We gave both models the same prompt: create a 3D animation in Blender with 900 frames at 30 frames per second, for a total of 30 seconds. Each model produced a clockwork mechanism filmed in macro. We rendered both animations on a single RTX 5060, at the same time, and then joined the results into one side-by-side video.

This was a hands-on test of one prompt, one run per model, not a controlled benchmark. Results can change with another prompt, another seed, or another round of feedback.

Watch them side by side

Claude Opus 5.5 Max is on the left. GPT-6.1 Sol Ultra code is on the right.

The test setup

Both renders ran at the same time on one GPU, so treat the times as a rough guide. They reflect how heavy each scene was to render, not how fast either model writes code.

What we noticed

These notes come from watching the video and studying still frames taken every two seconds. They describe what is on screen, not which result is better.

Framing. The left side stays in extreme macro. You see one or two parts at a time, such as a balance wheel, a gear cluster, or part of a large gear, and the subject changes noticeably across the 30 seconds. The right side shows a wider, elevated view of the whole movement on a round plate, and the camera drifts slowly closer and lower.

Depth of field. The left uses a very shallow focus, so a narrow band is sharp and much of the frame becomes soft orange and brown blur. On the right, most of the mechanism stays in focus, with blur mainly toward the edges and background.

Light and color. The left has a hazy, warm, low-contrast look in gold and amber, with small blue accents from screws and jewels. The right has higher contrast: copper against a dark grey plate, with directional light and cast shadows from the gears and pillars.

Materials. The left reads as smooth, polished brass, though fine detail shows only in the small in-focus areas. The right shows more surface variety, with weathered copper, steel arms, a glossy silver part, and a textured plate. Gear teeth, screws, and the coiled hairspring are readable across much of the frame.

What they share. Both show a watch-style movement with a balance wheel and hairspring on a warm orange-to-terracotta background. Both use slow camera movement with no cuts, and both include floating dust.

Spots to check as you watch. Some viewers will want a clear view of the whole mechanism, which the left side shows little of. Others will prefer a cinematic macro feel, which the right side may show less of, since it can read as a set on a table. Look for soft areas near the upper right of the right side, parts cropped by the frame edge on both sides, and how much of each balance wheel is in focus when it is the main subject.

Pricing and access

Anthropic lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, released September 22, 2026, with effort levels from Low to Max. OpenAI lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, released September 29, 2026, and says it is available in ChatGPT Work, Codex, and the API as gpt-6.1-sol. We did not record token counts or cost for these runs, so we make no claim about which was cheaper to produce.

How to judge it yourself

Different jobs reward different things, so decide what you are choosing for before you watch again:

• Brief fit: Which animation looks more like the macro clockwork shot you would have asked for?
• Readability: Can you follow how the mechanism works, or do you mostly feel it?
• Camera and pacing: Does the movement hold your attention for 30 seconds?
• Finish: Which scene would need less cleanup before you could use it?
• Time and cost: Is the extra render time worth what you see on screen?

If you repeat the experiment, keep the settings as matched as you can, render on separate runs for clean timings, and save the prompt so others can reproduce it.

Privacy note

Both Anthropic and OpenAI maintain public privacy policies, and data handling differs between consumer and business plans. Check the settings and terms that apply to your account before sending confidential scenes, assets, or briefs to either model.