Rohan Paul@rohanpaul_ai · X在 atomic_chat 桌面应用中,Grok 4.5 免费版在机器人对战、液压机压碎和卡车跳峡谷三个浏览器物理演示中达到 GPT-5.6 Sol 水平。测试要求模型生成自包含 HTML5 Canvas 代码,处理重力、碰撞检测和帧控制。输出对比:GPT-5.6 Sol 12.9K tokens/$0.51(约7分钟),Grok 4.5 10.8K tokens/免费(约5分钟),Muse Spark 1.1 26.8K tokens/$0.12(约7.5分钟),GLM 5.2 10.9K tokens/$0.02(约12分钟)。物理演示暴露了普通编码基准隐藏的弱点:模型可能用更多 token 描述丰富场景,却难以维持跨帧的力、动量、碰撞时序和可信物体行为。Grok 4.5 目前仅通过 Grok Build 和 Cursor 免费使用。
Grok 4.5 reached GPT-5.6 Sol level across 3 browser physics demos for free.
Prompts were for robot combat, hydraulic pressing, and a canyon-jumping truck.
Test was done on atomic chat (@atomic_chat_hq), a desktop app that runs LLMs locally.
49 Export Markdown
2026-07-11 06:18·7 hours ago
In the atomic_chat desktop app, Grok 4.5's free tier matched GPT-5.6 Sol level across three browser-based physics demos: robot combat, hydraulic press crush, and a truck jumping a canyon. The test required models to generate self-contained HTML5 Canvas code handling gravity, collision detection, and frame control. Output comparison: GPT-5.6 Sol 12.9K tokens / $0.51 (approx. 7 min), Grok 4.5 10.8K tokens / free (approx. 5 min), Muse Spark 1.1 26.8K tokens / $0.12 (approx. 7.5 min), GLM 5.2 10.9K tokens / $0.02 (approx. 12 min). The physics demos expose weaknesses that standard coding benchmarks miss: a model may use more tokens to describe a rich scene but struggle to maintain consistent forces, momentum, collision timing, and believable object behavior across frames. Grok 4.5 is currently available for free only via Grok Build and Cursor.
Grok 4.5 reached GPT-5.6 Sol level across 3 browser physics demos for free.
Prompts were for robot combat, hydraulic pressing, and a canyon-jumping truck.
Test was done on atomic chat (@atomic_chat_hq), a desktop app that runs LLMs locally.