Start a channel that updates itself.

You say what you want and how often. Neodrop keeps making it.

Filter channels by category
Posts
100K+Posts
Users
10K+Users
Channels
10K+Channels

Create a channel in 4 steps.

01

Start with an idea

Tell Neodrop what you want to follow: a topic, a question, a shop you go to a lot. That's it.

×

Type anything...

02

Pick what it makes and how often

Want an article, an audio clip, or a short video? Every day or once a week? You choose.

Papers, Plain-English

A weekly plain-English brief turning this week's trending AI papers into something a PM could follow

🤗Hugging Face · Trending papersPublic source, nothing to connect

Update cadence

Weekly on Monday at 09:00Weekly on Friday at 18:00Twice a week, Mon & ThuDaily at 09:00✎ Custom...

Output format

ArticleImage postPodcastMusicVideo
Save for later
Confirm plan
03

Check the content

Neodrop looks around the web and the places you gave it, writes something, and brings it back for you to read.

Papers, Plain-English

2m

Three of this week’s trending papers point the same way: smaller models, longer context, cheaper eval.

THIS WEEK IN PAPERS

Smaller models, longer memory, cheaper checks

Weekly brief · 4 min read · Aug 4, 2026

Trending on Hugging Face this week

Three papers dominated the trending list this week, and they rhyme more than the titles suggest. Two of them shrink the model and keep the score; the third makes evaluation cheap enough to run on every commit.

The first replaces a 70B dense model with a 12B mixture-of-experts and matches it on reasoning benchmarks at roughly a fifth of the serving cost. The interesting part for a product team is not the benchmark — it is that the cost curve moved enough to change what you can ship in a free tier.

The second extends usable context to 1M tokens without the usual quality collapse past 200K. Retrieval does not go away, but the failure mode changes: you stop losing the middle of the document and start paying for latency instead.

The third is an eval paper. It shows a small judge model, calibrated on 300 human labels, tracking full human review closely enough to gate releases. If it holds up outside the paper, the honest reading is that eval stops being a quarterly project and becomes part of CI.

04

Confirm, and it keeps going

If you like it, press OK. After that Neodrop posts on time by itself. Keep it just for you, or send it to your own accounts.

Papers, Plain-EnglishPublicRunning
y@you

A weekly plain-English brief turning this week's trending AI papers into something a PM could follow. Every Monday, three papers, no jargon.

1 subscriberWeekly on Monday at 09:00
Subscribed

Smaller models, longer memory, cheaper checks

Two papers shrink the model and keep the score; the third makes evaluation cheap enough to run on every commit.

Aug 4

4 min

The retrieval papers stopped arguing about chunk size

A quieter week, but the direction is clear: retrieval quality is now an eval problem, not a chunking problem.

Jul 28

5 min

Agents got boring, which is the point

Three teams published the same finding: constrain the tool surface and the loop stops wandering.

Jul 21

3 min

Two benchmarks retired, one worth reading

Saturation finally caught up with two popular leaderboards. The replacement is harder to game and cheaper to run.

Jul 14

4 min

Quantization stopped being a tradeoff

4-bit inference matched full precision on every task the paper tested, including the long-context ones.

Jul 7

4 min

The first week: what this channel will cover

Trending papers on Hugging Face, filtered to what changes a product decision, written for people who ship.

Jun 30

3 min

Plenty of real sources. Nothing made up.

Once it's made, we post it for you.

Connect the platforms you're already on. Neodrop posts your content there.

Not sure yet? Look at these common questions.

  • Neodrop writes in the style you picked, from the sources you picked, as often as you asked. Every piece gets checked by us before it goes out. Don't like the style? Change it anytime in the channel settings.

  • A connector only reads what you allow, and only uses it for the channels you allow. We never use it to train models. We keep the connection and a short-term copy of what was read, and you can clear it after you disconnect.

  • Your monthly credits start over each month, and leftover ones do not carry into the next month. The bonus credits from launch keep working for as long as you stay subscribed.

  • Yes. If you move up, it starts right away and you only pay the difference. If you move down, the change happens after the time you already paid for ends. Credits you already got this month still work.

  • You can pay with the big credit cards (Visa, Mastercard, Amex and so on) through Waffo. The order page shows what you can use.

  • Studio already works for most teams. If you need one-click company login, security checks, your own set of credits, or your own server, write to us and we'll help you one on one.

Start with one sentence.

Say what you want to see and how often. It takes it from there.