← Today's paper

Saturday, November 21, 2026

What this paper cost

Price$1.45

An estimate, at the model makers' published API prices.

The Fritter Post is written every morning by a program that reads the news and a handful of AI models that judge and write it. The price above represents the cost of the words read and written by those models for today's paper. It's based on API prices, not what I actually pay (I have a flat-rate subscription).

936articles read from 107 sources
139stories in today's paper
304model calls, 1.4 million tokens
21 minfrom first feed to front page

How the paper is made

Nine steps, every morning

Software does the mechanical work. The models do the reading, judging and writing. Here's what each step did today, and what it would have cost.

  1. 01

    Collect

    $0.00

    Checks 107 news sources for everything published since yesterday morning.

    Software only·936 articles·under 1 min

  2. 02

    Clean up

    $0.27

    Removes duplicates, stale stories and junk, and translates the articles that aren't in English.

    Qwen3.6 35B·59 calls·229K tokens·2 min

  3. 03

    Screen

    $0.22

    A first read of every article against what this reader cares about. Cuts what doesn't belong, and tells news from opinion.

    GLM-5.2·22 calls·134K tokens·1 min

  4. 04

    Group

    $0.09

    Finds articles about the same event, so ten outlets' reports of one fire become one story with ten sources.

    Qwen3 Embedding 8B·GLM-5.2·87 calls·131K tokens·2 min

  5. 05

    Score

    $0.31

    Judges every story for how much it matters and to whom, checks it is actually new, and links stories that belong to an ongoing situation.

    GLM-5.2·GLM-5.2 (thinking)·18 calls·150K tokens·5 min

  6. 06

    Rank

    $0.14

    Puts the paper in order by score and by how many separate outlets reported each story, and decides which get a full piece. A model is asked only to break ties.

    GLM-5.2 (thinking)·24 calls·50K tokens·1 min

  7. 07

    Read

    $0.00

    Fetches the full text behind the bigger stories, so they are written from the reporting and not from a summary.

    Software only·81 articles·under 1 min

  8. 08

    Write

    $0.39

    Writes every piece in the paper from its sources, in the house voice: features, standard stories and briefs.

    DeepSeek V4.1 Flash·93 calls·729K tokens·9 min

  9. 09

    Publish

    $0.02

    Lays out the paper, attaches the links to the original reporting, and notes when the paper has covered a story before.

    GLM-5.2·1 call·12K tokens·under 1 min

About these numbers

Every call counts, including retries and calls that failed, because a provider bills those too. The token counts are the ones each provider reported back.

Prices are list prices from each model's maker, with no caching or off-peak discounts, as of October 2026: z.ai, deepseek.com, alibabacloud.com, openrouter.ai.