I watched this voice interaction effect of DeepSeek several times.
What I like most is not the waveform itself, Instead, when going from "Speak" to "Swipe up to cancel", the status switches from blue to red. You can almost tell what you are doing without reading the text. When AI products come to the end, they may not only compete for model capabilities, but also And there are these little 0.5 seconds of interaction details.
I tested DeepSeek V4 Flash, Kimi K3, and Muse Spark 1.3 on the same Flappy Bird prompt.
Meta Muse Spark 1.3 was my favorite. It built the most complete game with a polished UI, animated bird, layered parallax, pipes, scoring, persistent best score, and a proper game-over
[90% don't know] For those who want to save 30,000 yen per month on AI I completely replaced my Opus 5 with a 1/20th cheap model (Deepseek V4 Flash)
The only weakness = not being able to see the image was solved by just adding 2 packages. People who pay 30,000 yen a month will lose money if they don't know this😳 ・$200 a month has been drastically reduced to 1/20th ・Only 2 commands are used
Hy4 preview just dropped, so I tested it on WorkBuddy against three other models
DeepSeek V4 Flash, GPT-5.6-Luna, and GLM-5.3... the top 4 models on this week’s OpenRouter coding leaderboard. The best part? They’re all available in WorkBuddy, so you don’t need to download
You have reached the end of the archive
All of deepseek