deepseek-v4.1VIDEO
同题写 Doom 游戏:DeepSeek V4.1 Flash 稳定出结果,Nanbeige 过度思考
在 2×3090 上约 40 tok/s 的 DeepSeek V4.1 Flash 1017 与单卡 5090 上 70 tok/s 的 Nanbeige4.3-33B 同做 Python+tmux 盲文渲染 Doom,前者持续产出,后者反复过度思考。
Aleksey Smolenchuk
@lxe2 天前收录原推发布于 2026.10.26
Aleksey Smolenchuk @lxe原推发布于 2026.10.26
Putting Nanbeige4.3-33B through the Doom gauntlet on a single RTX 5090. 2.5 bpw EXL3, bespoke Rust + CUDA runner: 70 tok/s Against DeepSeek-V4.1-Flash 1017 on 2× RTX 3090 at ~40 tok/s. The task: build and run a playable Doom-style game in Python, rendered with colored Braille in tmux. DeepSeek consistently produces results while Nanbeige consistently overthinks. I'm validating it against original 16-bit but so far it tends to overthink as well :/
x.com/lxe/status/2114458953329639485采集于 2026.10.26 12:02