Minerka The AI briefing

Library

Research

Papers, system cards, and technical reports. A product post, a system card, and a paper are different evidence.

Follow this shelf

Start here

One worked check, four list prices, then three questions — before you trust what changed this week.

Worked check: PlanAssist cut officer assessment time up to 58% in pilots — DeepMind’s figure; a human still decides. Field card · prices and 4 checks

Pocket list (November 1, 2026)

  1. Measurement — which named test, on which date, scored by whom.
  2. Bill — list input and output per million tokens on the vendor page.
  3. Door — live API, lab claim, replay, or a downloadable file — keep them separate.

Models · Agents · Fine-tuning · All chapters

In this briefing

  1. Artificial Analysis puts 60 percent of its index on private tests, and Opus 5.5 leads at 58Medium impact · October 27, 2026 · On October 27, 2026, Artificial Analysis replaced its Intelligence Index. Sixty percent of the new index uses private tests or held-out answers. Claude Opus 5.5 leads at 58. Artificial Analysis says the gap to the best open-weights model widened from 3 points to 12.
  2. Anthropic's partners report a 70 percent shorter design, and the drugs are not in people yetMedium impact · October 27, 2026 · On October 27, 2026, Anthropic said six drug partners have been using Claude. Generate Biomedicines reported one design about 70 percent shorter that, in lab tests, let immune cells make roughly 10 times as many receptors. Anthropic says the work is mostly preclinical.
  3. Ai2 finds a novelty judge can fall from 93.5 percent to 40.9 percent when the prompt changesMedium impact · October 27, 2026 · On October 27, 2026, the Allen Institute for AI released a study of models that score whether a research idea is new. For GPT-5.4, one prompt scored 93.5 percent on pairs of human and generated ideas. A shorter prompt on the same pairs scored 40.9 percent. A simple Claude Opus 4.6 prompt cost $2.59 and beat two purpose-built judges that cost $30.66 and $82.05.
  4. Epoch grades two leading models at 65 percent of its own work standardMedium impact · October 8, 2026 · Epoch AI scored models on tasks from its own work. Claude Fable 5.1 and GPT-6 Astra each averaged 65 percent of the rubric. Epoch says 100 percent is the line at which a task can be handed over, and that a 65 percent result is still a failed task.
  5. Carbon-A proposes 566 million gene candidates across 22,617 speciesMedium impact · October 8, 2026 · Hugging Face released a gene-finding model and a database of 566 million predicted protein-coding loci.

Key companies and projects

  • Hugging Face
  • Google DeepMind
  • OpenAI

Related tags

  • benchmarks
  • preprints
  • system-cards
  • Artificial Analysis
  • evaluation
  • research
  • Anthropic
  • Claude
  • Ai2
  • Epoch
  • agents
  • Hugging Face
  • biology

Three documents, three jobs

A product post tells you what shipped and who can use it. A system card tells you how the lab says it behaves under abuse and where it refuses. A paper or technical report tells you the method, the data, and the comparison. Carbon-A has a blog post, a Hub model, and a technical report. Odyssey-3, so far, is a lab post and a preview. Those are different evidence grades.

We do not treat an internal evaluation as a public leaderboard. OpenAI’s speed and quality figures for GPT-6 are labeled as internal. They can still be the right numbers to quote, with that label attached.

The questions we ask every time

What can a careful person check? What was held out of training? Who measured it, the authors or someone else? What would make the claim false in the next month? If we cannot answer the last one, the item stays short.