DEMO / RECORDING
Overview & result
The author compared an earlier GPT-6 Astra game-playing attempt with a new Jev run in Slay the Spire 2. In the post, they report roughly 0.7 seconds of action planning and share gameplay video. The post does not provide source code or a breakdown of how the game state is read.
Show the original post (Chinese)
昨天出的 https://t.co/6eiMFaJTrj Jev 模型
适合高频决策，那岂不又是适合打游戏？
之前用 GPT-6 Astra 代打杀戮尖塔2
能力强但速度慢。
我刚实测用 Jev 打，行动思考只需 0.7秒，
画面我都没看清它就操作完了。
这超人类游戏速度，
着实又让我震惊瘫坐了😅
语言很难描述这感觉，看视频⬇️ https://t.co/y9Nu78Hnv4
How Jev fits in the loop
- The linked demo shows an agent making game actions with Jev.
- The author reports about 0.7 seconds of thinking per action.
How to reproduce
- Watch the gameplay video in the original X post.
- No public repository or reproducible setup is linked from this case.
Why this build matters
It provides a visible example of Jev used for frequent decisions in a game.
Reported performance
Reported by authorLatency: ~0.7s / action (author claim)
The author says action planning with Jev took about 0.7 seconds; no cost or separate model latency is reported.
Limitations
- The post does not describe its game-state extraction, candidate-action generation or independent benchmark.