Qwen 3.8 27B Beats Claude Opus 4.6 in Three.js Coding Test for Free
testingcatalog · x · 2026-08-15
AI/ML API conducted a benchmark task: generating a single self-contained three.js file to render a sunken Atlantis scene (temple, columns, fish, kelp, caustics) with a scripted 20-second camera flythrough.
Results show that the open-weights Qwen 3.8 27B (self-hosted) outperformed Claude Opus 4.6 in code density and scene richness:
- Qwen generated a full temple colonnade, domed roof, and a centered glowing portal, with layered elements like fish and kelp.
- Opus produced a long row of columns fading into fog, with a small, off-center portal and a thinner scene.
Cost-wise, Qwen ran for free (self-hosted), while Opus cost $0.21 per run.
More from coding & agent
- Terminal Night 3 talk slides released, missing demo video to follow — 4310sy · 2026-10-16
- Agent developers are sitting on unused API credits that can exceed actual usage — EnvironmentalDay8864 · 2026-10-16
- Notion's bet on agents: abstract away deployment so users and agents just write code — simonguozirui · 2026-10-16
- Four Gemini 4 Argon agents beat Minecraft's Ender Dragon autonomously in 27 hours — Maxence Faldor · 2026-10-16
- Tencent's Terminal-RSI builds recursively self-improving terminal agents, hitting 19.84% on Terminal-Bench 4.0 — tencent · 2026-10-16
- Tsinghua's AutoSFT uses multi-agent collaboration to automate SFT data research — TsinghuaNLP · 2026-10-16