返回
原文
Rohan Paul@rohanpaul_ai
40AI 编辑部评分，满分 100
2026-08-03 20:50· 34分钟前
跳到正文
AI 摘要

Qwen 3.8 Max 在构建 3D 物理场景上击败 Fable 5，运行成本低约 7 倍。测试由本地运行 LLM 的桌面应用 Atomic Chat 完成：Qwen 3.8 Max 输出 46.6K tokens、花费 $0.28，Fable 5 输出 38.7K tokens、花费 $1.93。

Qwen 3.8 Max just built better 3D physics scenes than Fable 5 while costing about 7x less to run.

Test was done by @atomic_chat_hq ， a desktop app that runs LLMs locally.

Outputs： Qwen 3.8 Max： 46.6K tokens， $0.28 Fable 5： 38.7K tokens， $1.93

Both models got the same brief， three self-contained HTML files， each a scene that simulates itself.

The prompts asked for a marble machine lifting marbles onto a wheel and a loop， a car assembly line， and a sawmill cutting logs into planks.

These tests were actually pretty brutal. Shows whether the model maintained object state， collisions， and machine geometry across the whole scene.

atomic.chatQwen 3.8 Max beat Fable 5 at building 3D physics scenes for 7x cheaper! We gave two models the same task. Build three self-contained 3D scenes, each one HTML fi...
编码评测/基准
在 X 查看原推导出 Markdown
返回精选列表
40导出 Markdown
Rohan Paul · @rohanpaul_ai · X·2026-08-03 20:50·34分钟前
在 X 看原推· x.com
AI 摘要

Qwen 3.8 Max 在构建 3D 物理场景上击败 Fable 5，运行成本低约 7 倍。测试由本地运行 LLM 的桌面应用 Atomic Chat 完成：Qwen 3.8 Max 输出 46.6K tokens、花费 $0.28，Fable 5 输出 38.7K tokens、花费 $1.93。

Qwen 3.8 Max just built better 3D physics scenes than Fable 5 while costing about 7x less to run.

Test was done by @atomic_chat_hq ， a desktop app that runs LLMs locally.

Outputs： Qwen 3.8 Max： 46.6K tokens， $0.28 Fable 5： 38.7K tokens， $1.93

Both models got the same brief， three self-contained HTML files， each a scene that simulates itself.

The prompts asked for a marble machine lifting marbles onto a wheel and a loop， a car assembly line， and a sawmill cutting logs into planks.

These tests were actually pretty brutal. Shows whether the model maintained object state， collisions， and machine geometry across the whole scene.

atomic.chatQwen 3.8 Max beat Fable 5 at building 3D physics scenes for 7x cheaper! We gave two models the same task. Build three self-contained 3D scenes, each one HTML fi...
编码评测/基准
在 X 查看原推x.com