Daily Brief Oct 16, 2026
How we checked every item today except "Also happened"
For readers who want to dig in. What happened, in brief, is on the Today tab.
This issue draws on SecondSource's own research brief, produced early on October 16, and last night's sweep. The six main-line items date from October 14 to 15. Most items in the later sections date from October 13 to 14; the dated catch-ups and the archive pick are the exceptions. Last night's sweep brought in 732 pieces, and this issue uses 29 outside sources, each with a link to check.
1. [Today] (event date October 15) Meta releases Muse Spark Open, an 816-billion-parameter open-weight model, and model-service sellers with over US$20M in annual revenue must sign separately with Meta
Not proven yet： "most capable US open-weight model" is Meta's claim alone
How we checked
Read in full: Meta's three posts, the Hugging Face write-up and the license section of the model page. The license's other terms haven't been reviewed clause by clause. Every benchmark score is Meta's own, and no independent evaluation came out on release day. Neither the three posts nor the write-up says whether Spark Open comes from the same model generation as the closed Spark released in April. Apple's statement falls outside this issue's sources.
2. [Today] (event date October 15) TSMC's third-quarter gross margin reaches 66.4%, 2 nm takes 8% of wafer revenue in its first quarter, and fourth-quarter revenue guidance is US$52.3B to US$54.3B
Not proven yet： what drove the margin up; the press release gives no wafer prices
How we checked
The 6-K exhibit was read in full, with figures and quotes checked word for word. It is a statutory filing and a primary document. The earnings-call presentation filed alongside it hasn't been reviewed; the high-performance-computing share, full-year capital spending and 2 nm customers are all in that deck. Guidance is management's expectation.
3. [Today] (reported October 15) The Financial Times reports that Anthropic told investors distillation once made up 5% to 7% of revenue and was cut back sharply in the third quarter
Not proven yet： which period the 5% to 7% covers isn't in the quote, and the full story is unreviewed
How we checked
A single reporting chain resting on four anonymous sources. The basis here is Willison's verbatim excerpt; the Financial Times original is paywalled and hasn't been reviewed. Investor Kevin Xu says in a separate post that the 5% to 7% figures "are accurate". The basis for that is a rough estimate Kevin Xu made in June with an operator who once ran a distillation relay service. That is a personal estimate, not a second source.
4. [Today] (event date October 15) Anthropic gives six people from METR, a nonprofit that evaluates frontier-model risk, employee-like access; Anthropic can't veto METR's reports and can only require redactions in predefined categories
Not proven yet： the agreement text and METR's own account are outside this issue's sources
How we checked
The announcement was read in full, and all of it is the evaluated company's own account. The announcement says the arrangement runs under "enforceable agreements"; neither the agreement text nor any account from METR's side is in this issue's sources. The announcement gives no redaction deadlines or renewal terms, and doesn't spell out the tiers of the four other partners.
5. [Today] (post October 15) Investor Kevin Xu says the US data-center moratoriums he tracks now number more than 400, most of them pauses that expire within a year
Not proven yet： the counting method isn't in the post, and the dashboard hasn't been reviewed
How we checked
The post was read in full, with quotes checked word for word. It is a single source and the first reading from the author's own dashboard. The counting method isn't in the post, and the dashboard page hasn't been reviewed. "Raises GPU rental prices" comes with no rental-price data. The charts attached to the post haven't been read.
6. [Today] (event date October 14) Super Micro changes CFO: David Weigand steps down and former Western Digital CFO Mark Long takes over immediately, with no reason given in the filing
Not proven yet： the filing gives no reason for the change and doesn't say where Weigand goes
How we checked
The 8-K was read in full, with the quote checked word for word. It is a statutory filing and a primary document. The filing gives no reason for the change and doesn't say whether Weigand keeps another role. Recorded here without inference.
[This quarter] (arXiv preprint, October 2026) ETH Zürich and NYU's SAFARI group test 36 HBM2 chips and find an access pattern that draws 23.6% more power than the highest-power measurement loop in the JEDEC memory standard, averaged across the 36 chips.
How we checked
The abstract doesn't say whose chips the 36 are or at what temperature and frequency they were measured, and the full paper hasn't been reviewed.
[This quarter] (released September 30) PyTorch 2.14.1 moves to CUDA 13.2.2 and fixes a cuBLAS bug that could ignore the tensor-wide scaling factor in NVFP4 matrix multiplications.
How we checked
(PyTorch 2.14.1 release notes, 2026-09-30) (PyTorch 2.14.0 release notes, 2026-09-02)
[Today] (post October 14) Buy-side researcher Rihard Jarc says that when credit tightens, the bottleneck becomes capital, existing compute gets scarcer, and compute prices may rise rather than fall.
How we checked
[This quarter] (released September 3) Alibaba's Qwen team and Taobao/Tmall publish E-Commerce Bench, in which 18 models each run a simulated online shop for a year. Only one model clearly learns to push down repeat purchase prices, and it is Qwen's own Qwen3.8-Max-Preview. GPT-5.6 Sol finishes first on year-end assets but ranks 16th on fraud defense.
How we checked
[This week] (preprint October 13) An arXiv paper uses randomized on/off trials to screen the lesson playbooks agents write: screened playbooks average a success rate of 66.9, against 54.1 with no playbook, 55.5 using the whole playbook and 53.8 letting the agent choose.
How we checked
The method, RBC, randomly switches each lesson in the playbook on or off and estimates each lesson's effect from the difference in success rates.
(arXiv 2610.18427) (arXiv 2610.18216; arXiv 2610.18813; arXiv 2610.18845)
[This week] (model card updated October 14; model released August 25) Z.ai, the developer of the GLM models, updates the self-reported benchmark table on the GLM-5.3 model card: GLM-5.3 leads on only three of 16 tests, and scores 54.4 on ExploitBench, an exploit-writing test, against 78.0 for Fable 5.
How we checked
These are one model's self-reported scores and say nothing about open-weight models as a class.
[Business] (case page undated; read on October 15) OpenAI customer story: LegalOn, a legal-software company, routes tasks for the coding agent Codex across three GPT-6 models and adds mode restrictions and budget caps, estimating daily costs about 65% lower than before.
How we checked
The page says: "Together, model selection, feature restrictions, and budgets tailored to each business reduced estimated daily costs by approximately 65%." It is the customer's own estimate, published by the vendor, and the measurement period isn't stated.
[Looking back] (deep dive, September 6, 2026) "How long will this security gate hold?" has no single answer; it depends on who is being defended against.
How we checked
Read with Model watch item 3: the GLM-5.3 model card's self-reported scores, close to closed flagships at finding vulnerabilities and still behind at writing exploits, are one more reading on the same list waiting to be checked. A model card's self-reported scores and the UK-US government evaluation are not the same yardstick.
Still missing today. What would most affect judgment is the full Financial Times story on Anthropic: it is paywalled, and which period the 5% to 7% covers isn't in the quote. Beyond that, the other terms of Meta's license, TSMC's earnings-call presentation, the agreement text and METR's own account, the counting method behind Kevin Xu's dashboard, Citi's CPU-market report, Liquid's quote page, and the full HBM2 and RBC papers haven't been reviewed.
Sources & accounting
The past 24 hours. 732 new pieces came in from October 15 to 16: 449 arXiv academic papers, 146 social-platform posts, 79 company or personal blog posts, 50 papers from other sources, 6 subscription newsletters and 2 company filings. Of those 732, 15 were read in full and 4 were screened out as off-topic; the other 713 haven't been read yet and remain on the reading list. All 449 arXiv papers are among the unread. The night before, arXiv's servers were overloaded and nothing was fetched, so these 449 cover two nights. The rest of the unread are 143 posts, 66 blog posts, 50 papers from one paper-digest subscription and 5 newsletters. Every arXiv preprint cited in this issue's later sections was read only as an abstract, none is peer-reviewed, and most don't name what they compare against. No new show transcripts came in last night. Four sources are logged as unfetchable: Data Center Dynamics, the Meta AI blog, and the investor news pages of Broadcom and Qualcomm; any announcements from them that evening fall outside this issue's sources. The TSMC filing, the Meta model page and license, and the Super Micro filing behind the main line were originals read directly today.
One-time catch-up. No new one-time catch-up of older material today.
A note on source concentration. ⚠️ The six main-line items each rest on independent sources, with none shared. ⚠️ The full reports or stories carrying main-line item 3 and "Also happened" items 1 to 4 haven't been reviewed. ⚠️ The scores and cost figures in all three Model watch items and the Product moves item are the vendor's or customer's own.
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 732 new pieces above; they can't be added or compared. Representative names: on social platforms, Mark Zuckerberg, Lucas Beyer, Graham Neubig, Arvind Narayanan and Jordan Schneider; among newsletters, Zvi Mowshowitz, Miles Brundage, Dean Ball, Kevin Xu and Ian Cutress; among shows, interviews with Sam Altman, Dario Amodei, Demis Hassabis and others. This issue uses 29 outside sources in the body. Only body links actually cited, and not on SecondSource's own domain, are counted; that is a different population from the 529 tracked sources above. On the Taiwanese side, 7 new sources were added in September, bringing the total to 128; Taiwanese media, Taiwanese companies' official channels and local reporters in Taiwan count, while foreign outlets relaying Taiwanese media do not.