Fable 5.2 leads on most tests but costs twice as much. When is the stronger model worth it? Also: Notion's AI chat breach.
From our library · 2026-11-10
Anthropic released Claude Fable 5.2 on November 9. In Anthropic's own tests, it beats its older model, Opus 5.5, on 11 of 12 measures. It also costs twice as much for each unit of text it reads or writes. The idea to learn is routing: send hard, long jobs to the stronger model and routine jobs to the cheaper one. Anthropic says Fable matched Opus on one coding test at about half the cost per task. A separate test found Fable spent far more on one drawing task. All scores come from the lab.
Why it matters: The better buy is the model that finishes your task for less money, not the one with the top score or the lower price tag.
What you can do: Run the same 20 to 50 real tasks through both models. Compare how many each one solves and what each costs before you switch.
Read the sourceNews · 2026-11-10
Notion told some customers that data taken from its AI vendor, Fireworks, included Notion AI chat transcripts. Notion says fewer than 1% of workspaces may be affected and that no other workspace data was. The details come from a customer email posted online. Notion has not published its own notice yet.
Why it matters: If anyone pasted a password or key into an AI chat, treat it as exposed and change it. The weak spot was a vendor, not Notion itself.
Read the sourceNews · 2026-11-08
A GitHub project called llama-modchip says it can run DeepSeek-V4-Flash, a very large model, on a 64 GB desktop with no graphics card. Its author reports about 9 to 11 tokens per second, where a token is a small chunk of text. Standard tools cannot open its files.
Why it matters: Running big models on hardware you own could mean less reliance on cloud services. These are one maintainer's numbers, so wait for independent tests.
Read the sourceNews · 2026-11-10
Trading firm Jane Street released agent_skills on GitHub under the MIT license, which lets anyone reuse it. Its first skill, socratic-tutor, makes an AI teach by asking questions, testing you first and tracking your level from 0 to 5. It starts only when you ask to be taught.
Why it matters: A skill is a plain text file of instructions, so you can read exactly how it behaves. This one is built not to flatter you.
Read the source
From us
A beginner workshop over two live evenings, with no coding needed. You practise building a Claude Project loaded with your documents and tone, writing an agent skill your team can install, making a prompt library and connecting Claude to the apps you use.
See the details
From us
Two live evenings for beginner to intermediate learners. You explore Microsoft 365 Copilot, use it inside Excel and other Microsoft apps, put Copilot Cowork to work, then build automations in Copilot Studio.
See the detailsOne vote per browser. An informal reader poll, not a representative survey.
Comments
Join in with your explainx community account. Comments are plain text (no links, up to 600 characters), follow the community guidelines, and are reviewed before they appear.
No comments yet. What did you think of this edition?
Loading…