Rocket Launch: two Groks, two local models

The setup, the caveats, the limits.

Prompt, verbatim: "Create a single self-contained HTML file using Three.js that renders a visually spectacular, bright, colorful, and highly detailed 3D animated rocket launch scene with powerful engine fire, thick smoke, dust, and dramatic environmental effects. Requirements: * Use Three.js loaded from a CDN. * Everything must be contained in one HTML file and work directly when opened in a modern browser. * Fill the entire browser viewport edge to edge with the 3D scene. Do not leave margins, borders, blank space, or unused areas. * Create a bright, polished, AAA-quality stylized 3D visual style with vibrant colors, detailed materials, cinematic lighting, soft shadows, atmospheric depth, and excellent visibility. * Create a large detailed rocket launch site surrounded by colorful terrain, launch structures, support towers, pipes, platforms, service equipment, barriers, and distant hills. * Make the rocket the clear central focus of the scene, large enough to see important details such as the body, nose cone, engines, fins, and surface panels. * Position the rocket vertically on a detailed launch pad at the beginning of the animation. * Include a launch tower or support structure beside the rocket with mechanical arms or clamps holding it in place. * Start with a brief calm pre-launch moment with subtle atmospheric movement such as light smoke drifting and small warning lights glowing. * Show the rocket engines gradually igniting beneath the rocket. * Create bright, powerful animated engine flames with multiple layers of fire, including a bright core, glowing exhaust, and longer flame trails. * Add intense heat glow and warm light from the engines reflecting onto the rocket body, launch platform, and nearby structures. * Create large volumes of thick animated smoke and dust spreading outward across the ground as the engines increase thrust. * Make the smoke roll, expand, swirl, and rise naturally instead of appearing as simple static clouds. * Add smaller particles, sparks, glowing embers, dust, and exhaust fragments around the launch area. * Make nearby grass, dust, loose objects, and lightweight vegetation react to the powerful engine exhaust. * As engine thrust increases, make the rocket subtly vibrate before liftoff. * Animate launch clamps or support arms releasing the rocket just before takeoff. * Make the rocket rise vertically from the launch pad with smooth but powerful acceleration. * As the rocket rises, create a long, bright exhaust plume extending downward. * Make the exhaust plume dynamically change length and shape as the rocket gains altitude. * Show smoke and dust continuing to spread around the launch pad after liftoff. * Add visible atmospheric distortion or heat shimmer near the engine exhaust if possible. * Include subtle camera vibration and a brief cinematic camera shake during ignition and liftoff without making the scene difficult to watch. * Use a smooth cinematic camera that follows the rocket upward while keeping the rocket and major launch effects clearly visible. * As the rocket climbs, smoothly reveal more of the colorful landscape, launch site, clouds, and distant horizon below. * Keep the rocket large and prominent throughout the sequence. * Add layered clouds at different heights, with the rocket eventually climbing toward and partially through them. * Use a bright blue daytime sky with colorful white clouds and strong sunlight so the rocket, flames, smoke, and environment remain clearly visible. * Avoid a dark night scene. The result should be bright, vibrant, high-contrast, and visually clear. * Use realistic-looking shadows and dynamic lighting, including the changing influence of the bright engine fire. * Keep the viewport visually dense and impressive throughout the animation, with no excessive empty sky or unused space. * Use layered terrain, launch structures, smoke, clouds, and atmospheric effects to maintain visual interest as the rocket ascends. * The animation should follow a clear sequence: rocket on launch pad → pre-launch smoke → engines ignite → flames intensify → smoke and dust rapidly expand → rocket vibrates → clamps release → rocket lifts off → rocket accelerates upward → long exhaust plume follows → camera tracks rocket toward the clouds. * Use smooth animation for the rocket, flames, exhaust, smoke, dust, sparks, particles, vegetation, clouds, lighting, shadows, and camera movement. * Do not add overlay text, titles, labels, HUD, captions, menus, buttons, borders, or UI elements over the scene. * The browser viewport should contain only the animated 3D rocket launch scene, filling the entire screen edge to edge. * Start the animation automatically when the page loads. * Add smooth camera controls only if they do not obstruct the scene or interfere with the cinematic composition. * Optimize for smooth performance in a modern browser. * No build process is allowed, and there must be no external dependencies other than the Three.js CDN."

Grok 4.6 and 4.7 were called through the xAI API at default reasoning with no token cap. 4.6 returned a complete file in 635 s. 4.7 twice wrote the entire file inside its reasoning and then answered with a single instruction to save it as .html — no HTML in the response. Those two runs are archived as DNF; nothing was extracted from the truncated reasoning summaries. The 4.7 arm shown here is the same prompt pasted into grok.com chat, which did deliver. It therefore has no token or latency receipt and is not comparable on cost — only on the artifact.

GLM-5.3-Flash (EXL3, TP2) and MiMo-V2.6-Flash (MXFP4, vLLM TP2 using tonyd2wild's DGX Spark recipe) ran the identical prompt with thinking on and no output cap. Both completed first try with zero console errors. Token counts are nearly identical (79,607 vs 80,343); MiMo's wall time was 1.9× shorter. Disclosed: MiMo's first attempt was discarded because the checkpoint's generation_config caps output at 2,048 tokens by default; the harness now sends an explicit max_tokens and the run shown is the uncapped rerun.

Recording method: 24 s fixed-step capture, no interaction (the brief specifies autoplay with no UI). All four originals are published unchanged.