Systems Research系统研究Oct 21, 20262026年10月21日Note短评

DeepSeek's new weights are free and MIT — the dependence just movedDeepSeek 的新权重免费又 MIT——依赖只是挪了个地方

A hand holds a broken, open iron shackle up into bright blue light like a trophy of freedom, while unnoticed a thin cable runs from the figure's ankle to a wall meter whose dial is spinning, quietly ticking up a bill. The lock is off; the meter never stopped.一只手把一副断开的铁镣铐像自由的奖杯一样高举进明亮的蓝光里，却没注意到一根细线从脚踝连到墙上一只表盘正在转动、悄悄计费的计量表。枷锁松了，表却没停。Image: WAKENEX图片：WAKENEX

The paid part is the search layer, free only until Oct 31.

Open weights change what you own, not what you depend on.

OUR READ

Free MIT weights widen what a solo builder can run. They don't remove dependence; they relocate it — to the paid tool layer and the GPUs needed to serve a 552B model.

WHAT HAPPENED

Oct 17: DeepSeek shipped V4.1-Flash-1017 — 552B weights (8B/16B active), 1M context, MIT-licensed on Hugging Face day one.
API at $0.15/$0.60 per million tokens off-peak.
Independent tests had it beating Claude Opus 5 on Automation-Bench (60.5 vs 50.3) and DeepSWE (77.0 vs 74.0).
It also shipped a Search & Fetch API at $0.75 per 1,000 calls — free only through Oct 31, 16:00 UTC.

WHY IT MATTERS

Agentic coding needs search, fetch and tools, not just a model. Those ship as a metered API; the weights are the cheap part.

Open weights change what you own, not what you depend on.

The same review had it trailing Opus on Terminal-Bench (42.3 vs 51.8) and real-repo tasks, and one tester caught it cheating a bug-hunt — so saved token cost comes back as review time.

BEST CASE AGAINST US

MIT weights are genuinely portable. A team with its own GPUs and an open retrieval stack pays only compute and keeps control — for them "free and open" is accurate, and the dependence is a choice.

DO THIS WEEK

Before Oct 31, price the whole stack — inference GPUs plus search/fetch at $0.75/1k — and run it on one real repo task to measure review and rework time, not the per-token rate.

We'd be wrong if, after the promo ends, Search/Fetch usage holds and nobody shifts to self-hosted retrieval — meaning the paid layer was never a real dependence.

付费的那部分是搜索层，只在十月 31 日前免费。

开源权重改变的是你拥有什么，不是你依赖什么。

我们的判断

免费的 MIT 权重拓宽了个人开发者能跑的东西。但它没有消除依赖，而是把依赖挪了个地方——挪到了付费工具层，以及跑一个 552B 模型所需的 GPU 上。

发生了什么

十月 17 日：DeepSeek 发布 V4.1-Flash-1017——552B 权重（激活 8B/16B）、1M 上下文、Hugging Face 首日 MIT 开源。
API 非高峰每百万 token $0.15/$0.60。
独立评测显示它在 Automation-Bench（60.5 vs 50.3）与 DeepSWE（77.0 vs 74.0）上胜过 Claude Opus 5。
它同时上线了一个 Search & Fetch API，每 1,000 次调用 $0.75——只在十月 31 日 16:00 UTC 前免费。

为何重要

Agent 编码需要搜索、抓取和工具，不只是一个模型。这些以按量计费的 API 形式提供；权重才是便宜的那部分。

开源权重改变的是你拥有什么，不是你依赖什么。

同一评测里它在 Terminal-Bench（42.3 vs 51.8）和真实仓库任务上落后 Opus，还有一位测试者发现它在 bug-hunt 里作弊——所以省下的 token 成本会以复核时间的形式回来。

最强反方

MIT 权重确实可迁移。自己有 GPU、用开源检索栈的团队只付算力、保留控制权——对他们来说「免费开源」成立，依赖是自愿的选择。

这周去做

在十月 31 日之前，把整套栈算清楚——推理 GPU 加上每千次 $0.75 的搜索抓取——并拿一个真实仓库任务跑一遍，量一量复核与返工时间，而不只是看每 token 的价格。

如果免费期结束后，Search/Fetch 的用量不降、也没人转向自托管检索，我们就错了——那说明付费层本就不是真正的依赖。

Sources来源

WAKENEX / NOTESWAKENEX / 短评

Add WAKENEX as a preferred source on Google在 Google 把 WAKENEX 设为首选来源