Hardware-launch landing pages
A shared task compares M3.1 Flash and Opus 5.5 layouts, feature cards and product details.
Recently updated / ASIA/SHANGHAI
CHEER LIN / ARCHIVE
93 distinct demos from posts published since October 19, 2026, 00:00 China time, with >50 likes and native images or videos. Model attribution and results are poster claims and have not been independently reproduced. Collection cutoff: 10/26/2026, 21:09 (China time).
93 results
A shared task compares M3.1 Flash and Opus 5.5 layouts, feature cards and product details.
Split-screen device models from M3.1 Flash and DeepSeek. Timing, throughput and subscription usage are the author's reports.
A split-screen plush-spider demo attributed to M3.1 Flash and GPT-6 Astra, with color changes and rotation.
Side-by-side campfire scenes from M3.1 Flash and DeepSeek. The post lists timing, tokens and cost; billing was not verified.
Pixel-city scenes from the same prompt for M3.1 Flash and DeepSeek. Preferences, timing and costs remain the author's reports.
Two models create miniature rinks with lighting changes. The author prefers MiniMax's night scene; quality and costs are unverified.
The task requests fish responding to food rather than looping animation. The pond and button are visible; full behavior code was not verified.
The author used Claude Code as the harness for a 3D whale-breach scene. The video contains several models' results.
A split-screen Kimi K3 and M3.1 Flash pond demo. WebGPU is claimed; the actual runtime backend was not independently verified.
The author attributes the rotating pagoda scene to a direct M3.1 Flash API call at max effort without a harness.
img2interior converts a bedroom reference into a scene. The recording shows interior details; the full editing scope was not tested.
Atomic Chat shows M3.1 Flash and DeepSeek road chases. This is a provider demonstration; costs are unverified.
In Atomic Fusion, local Qwen plans and M3.1 codes. The platform compares modes and reports its own speed measurements.
Four code-based visual scenes. They are presented as graphics demos, not validated real-world physical predictions.
The video labels animated kitchen scenes M3.1 Flash and Opus 5.5. Generation steps and other media tools are undisclosed.
The author iterates in MiniMax Code using reported subagents and frame checks. Toy-like rockets and geometric transitions are visible.
The platform labels rocket concepts M3.1 Flash and Opus 5.5. This does not establish direct video-pixel generation by either language model.
Four models tackle the same driving task. The author's view that MiniMax looks weakest is retained as counterevidence.
The author supplied a prompt and reference images. The video shows an ocean, grass, paths and character-view movement.
The author shows keyboard, SaaS and 3D scenes, explaining reuse of one keyboard render across assets. Counted as one workflow collection.
The video compares ticket pages and pixel scenes. Its city segment overlaps another case; this card counts the montage once for its additional webpage result.
Side-by-side rocket launches from M3.1 Flash and DeepSeek. Assessments of flames, quality and cost are the author's.
Two 3D pianos with squishy keys, bubbles and tempo controls. Audio and all interactions were not independently tested.
Higgsfield labels the workflow M3.1 Flash. Headphones and lighting are shown; underlying generation-tool roles are undisclosed.
Showing 24 / 93