Daily Brief Oct 14, 2026
How we checked every item today except "Also happened"
For readers who want to dig in. If you only want what happened, go back to Today.
This issue draws on the research brief we write ourselves, produced early on October 14, and last night's sweep. All three main-line items happened on October 13, as did most items in the later sections; the exceptions are the "From the archive" item and the podcast in "Named commentary," whose transcript arrived last night. We swept 185 pieces overnight, and this issue uses 24 outside sources, each with a link to check.
1. [Today] (event date October 13) Anthropic says daily code merges rose more than tenfold in four months to over 30,000, and version control and review tools it built itself have replaced Git
Not proven yet： the merge count comes only from Anthropic, with no outside measurement
How we checked
we read the Anthropic and Google posts in the original and checked the quotes word for word; for Amp we read an archived copy of the post. All three are companies describing their own work. GitLab CEO Bill Staples cited the same merge figures the same day, but the source is Anthropic's post, so that doesn't count as a second source (Bill Staples, 2026-10-13). Google's changes are public. We didn't rerun them. What would prove this wrong: by the first half of 2027, no second frontier lab or large software company has publicly rebuilt its version control or review layer. That deadline is one we set ourselves. Or a team that does no human review publishes a defect rate no worse than teams that do.
2. [Today] (event date October 13) Google opens Gemini 4 Argon's paid API at an introductory US$2 per million input tokens and US$10 per million output tokens, with prices doubling after November 30
Not proven yet： the Vals Index first place is Google's own claim, and we haven't read the leaderboard
How we checked
we read DeepMind's launch post in the original and checked the pricing and ranking word for word. The outside readings come from archived copies of Shipper's and Mollick's posts. We haven't seen Shipper's task set or scoring method. Mollick reports having used the model for some time before launch. Each has a single test, and no third party has rerun either. We didn't read the Vals Index leaderboard page.
3. [Today] (event date October 13) NVIDIA says it is open-sourcing all of Run:ai, the GPU scheduling software it bought in 2024, with no paid subscription, and the GPU cloud Lambda says it will integrate it
Not proven yet： only the KAI Scheduler is explicitly placed under the Apache 2.0 license
How we checked
we read the NVIDIA and Lambda posts in the original and checked the quotes word for word. The two are independent of each other, but both are vendors describing their own plans, and Lambda is an NVIDIA partner. There are no figures on paying customers converting or on usage. The only component the post explicitly puts under Apache 2.0 is the KAI Scheduler.
[Today] (October 13) Marvell, Microsoft and Oracle formed a working group at the open-hardware group OCP to write an open spec for a hardware security module built into the server, with a first draft targeted around this week's OCP summit.
How we checked
[Today] (post October 13) Independent semiconductor analyst Sravan Kundojjala says ASML's trailing four-quarter order fill rate fell from 126.8% in the second quarter of 2025 to 87.3% in the fourth, because bookings came in faster than they were recognized, not because demand weakened.
How we checked
We didn't check them against ASML's financial reports, and no quarterly breakdown was attached.
[Today] (post October 13) Tim Culpan, a former Bloomberg columnist who founded the publication Culpium, says CASCI, the index Culpan built to track revenue across Taiwan's AI compute supply chain, fell in September for the first time in 12 months because of capacity constraints.
How we checked
Its method isn't public, and the post doesn't give the size of the drop.
[Today] (podcast transcript, obtained by us the night of October 13) Amin Vahdat, who runs Google's infrastructure, says building 1 GW of power capacity at 99.99% reliability means building 2 GW; Vahdat also says repurposing retired training chips for inference isn't necessarily cheaper.
How we checked
[Today] (October 13) DeepMind's safety post says that after an RL run accidentally penalized shortcuts in the chain of thought, the model discussed shortcuts openly 61% less often, but the share of shortcuts it took silently rose from 11.2% to 36.4%.
How we checked
The passage we read gives no total count, so it doesn't show whether the absolute number of silent shortcuts rose. This is DeepMind's own testing, with no outside replication, and we didn't read the safety report PDF.
[Today] (October 13) Google DeepMind reports that Gemini 4 Argon scores 87% on ERQA, an embodied-reasoning benchmark, "outperforming every other frontier model," and that more than 4,500 data items it synthesized from one tabletop photo raised a real robot's success rate from 56% to 75%.
How we checked
No third party has rerun the cross-model comparison, and the post doesn't give the number of trials. The interview and the demo are both self-reported, so we keep both views on record.
[Today] (post October 13) Shopify launched a shop skill for AI agents: once a buyer approves in the Shop app, the agent receives a single-use payment credential and can check out on any website.
How we checked
The thing to watch is whether any third-party agent publicly connects to the skill.
[Looking back] (deep dive, September 14, 2026) The ways companies hold back frontier models fit in one table: one axis is which layer is held back, the other is how release happens, and release comes in only two forms, on a later date or now but only to a vetted list.
How we checked
This round we did not recheck whether the test the deep dive set for being wrong has been met.
What you are not getting today. What would most affect judgment is the three companies' defect or rollback rates for agent-written code: none has published one. Beyond that, we didn't read Every's task set, the Vals Index leaderboard, DeepMind's safety report PDF, the full CNBC and The Information stories, CASCI's methodology, or ASML's quarterly order breakdown.
Sources & accounting
The past 24 hours. 185 new pieces came in from October 13 to 14: 137 social-platform posts, 40 company or personal blog posts, 4 subscription newsletters, 3 show transcripts and 1 paper from other sources; no arXiv academic papers came in during this window. Of those 185, we read 75 and screened out 61 as off-topic. The other 47 haven't been read yet, and 2 posts came in with no content captured. Of the 47 unread, 25 are social-platform posts, 21 are blog posts and 1 is a paper. The 3 show transcripts in this window are two episodes of Sequoia's Training Data (Amin Vahdat; David Silver and Lasse Espeholt) and one of Latent Space, and we read all three; nine YouTube channels had no new episodes fetched. Five more sources failed to fetch last night: Data Center Dynamics, DeepSeek's official blog, the Thinking Machines blog, and the investor news pages of Broadcom and Qualcomm. If any of them posted something new that night, we didn't receive it. So when we say something did not appear in our sweep, that does not cover the 47 unread pieces, the 2 empty posts or the five sources that failed to fetch. The Anthropic, Google and NVIDIA posts behind the main line, and the Lambda blog post, are originals we read directly today.
One-time catch-up. No new one-time catch-up of older material today.
A note on source concentration. ⚠️ In all three main-line items and both Model watch items, the key source is the blog of the company being reported on; Google's first-party developer blog and DeepMind posts make Google the company that supplies the most key sources in this issue. ⚠️ Both outside readings in main-line item 2 come from posts on X, each from a single person or organization. ⚠️ Chips items 2 and 3 each rest on figures one analyst calculated.
The sources we track. Our long-term roster has 529 named sources: 302 on social platforms, 90 shows, 51 news outlets, 48 blogs, 48 paper authors and 46 newsletters, with the rest spread across earnings reports, keynotes and other channels. ⚠️ Those are counts of tracked sources, a different population from the 185 new pieces above; they can't be added or compared. Representative names: on social platforms, Mark Zuckerberg, Lucas Beyer, Graham Neubig, Arvind Narayanan and Pushmeet Kohli; among newsletters, Zvi Mowshowitz, Dean Ball, Jordan Schneider and Ian Cutress; among shows, interviews with Sam Altman, Dario Amodei and Demis Hassabis. This issue uses 24 outside sources in the body. We count only body links not on our own domain, a different population from the 529 tracked sources above. We added 7 Taiwanese sources in September, bringing the Taiwanese total to 127; Taiwanese media, company official channels in Taiwan and local reporters there count; foreign outlets relaying Taiwanese media do not.