WHAT NEW IN X10.30 FRI

X 上的 AI 合集档案 · 22 个合集 · 60,637 条

deepseek-v4.1VIDEO

DeepSeek V4.1 Flash 接入 Claude Code:成本降 13 倍

通过 Anthropic 兼容端点三行环境变量把 Claude Code 切到 deepseek-flash[1m],输入/输出 $0.30/$1.20 每百万 token,周末半价;DeepSWE 77 胜 Opus 5 的 74,但 Terminal Bench 42 落后 52。

Findr
@aifindr11 天前收录原推发布于 2026.10.18
Findr @aifindr原推发布于 2026.10.18
在 X 上打开原推

DEEPSEEK JUST MADE CLAUDE CODE 13X CHEAPER. 26X ON WEEKENDS. same terminal. same workflow. you just swap opus for v4.1 flash 1017. the part the launch posts skip: deepseek runs an anthropic compatible endpoint. 3 lines: export ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic export ANTHROPIC_AUTH_TOKEN=your_deepseek_key export ANTHROPIC_MODEL="deepseek-flash[1m]" then run claude like always. the math per 1M tokens, input / output: > opus 5.5: $4 / $20 > deepseek flash: $0.30 / $1.20 > off peak and weekends: $0.15 / $0.60 why it's that cheap: every layer has 384 experts. only 6 wake up per token. the other 378 sleep. that's the whale in the video. the catch, straight from deepseek's own table: > beats opus 5 on DeepSWE, 77 vs 74 > loses to opus 5 on Terminal Bench, 42 vs 52 so keep opus for the hard calls. let the whale do the grind. your claude code bill is a setting, not a law.

这条是引用着另一条写的:x.com/juliangoldieseo/status/2111525607763034116

x.com/aifindr/status/2111535661593559376采集于 2026.10.18 16:35