Friday, November 6, 2026 Edition no. 13 · Morning
PTENES
Noticiário do Joãozinho Search

Artificial intelligence news from Brazil, in English

Noticiário do Joãozinho: Edition no. 13 · Nov. 6, 2026 · Morning

Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese

Top storyEdition no. 13 · Nov. 6, 2026 · Morning

Thinking Machines Lab launches Inkling 2 Small: MoE 594B/17B active, AA Index 31; Apache 2.0 weights next week

On 11/05/2026 Thinking Machines Lab (CEO Mira Murati) announced Inkling 2 Small, the second generation of the Inkling family: a Mixture-of-Experts with 594B total parameters and 17B active, with image input and variable effort. On the Artificial Analysis Intelligence Index it scores 31 (post table: 30.6 on AA Index v5.0). Available today on Tinker (Playground, fine-tuning and sampling) with closed weights; the lab says it intends to release the full weights under Apache 2.0 next week after external safety testing and red-teaming. Pre-training H1: 8.4T tokens in 68h on NVIDIA GB300 NVL72 racks in NVFP4; H2 3.8T at 512K; then H3/H4 phases and a joint RL run of 1,075 steps / 17M+ rollouts / 9 days on 2,048 GPUs (>4×10²³ FLOPs, 3T+ tokens). Vision encoder 292M (Perception Encoder). Highlights in coding/vision/finance (e.g. DeepSWE 72.1%; DesignArena Website 1,340; AA-Briefcase 1527). The official @thinkymachines X post had about 736 thousand views and 3,581 likes in this edition's measurement (~8 a.m. BRT on 11/06).

Sources

PDF (pt-BR)

Models

1 story

High importanceNov. 5, 2026

Ant Ling launches Ling-3.2-flash: native multimodal MoE 560B/25B active, 1M context

On 11/05/2026 Ant Ling launched Ling-3.2-flash (official changelog + site + X @AntLingAGI): a native multimodal MoE with 560B total parameters and 25B active, a 1M-token window, native support for image, text and video. Focus on 4 axes: complex visual understanding, multimodal long-horizon Agent, 3D understanding & interaction, medical multimodal. Available in chat and via an OpenAI/Anthropic-compatible API; free for the first 2 weeks. On X, the lab claims to beat Opus 5 (max) on OmniDocBench1.5 (92.21 vs 90.56), ZEROBench Sub (43.71 vs 39.52) and Vision2Web (85.86 vs 80.20).

Source Ant LingModel docsSiteX announcement @AntLingAGI

Read the story

Research

1 story

Medium importanceNov. 6, 2026

Science's Last Exam opens 113 public tasks (11/06); FrontierPhysics: Opus 5.5 ~27.4% Pass@1

Science's Last Exam (SLE), led by the AItonomy Foundation with researchers from 34 institutions (Harvard, MIT, Stanford, Oxford, Princeton, ORNL etc.), announces its official launch on 11/06/2026: open-sourcing 113 public tasks across 9 disciplines, covering 6 capabilities + end-to-end research. Public downloads open on 11/06. In parallel, the FrontierPhysics preprint (arXiv 2611.06910, submitted ~11/04/2026) evaluates agents on 56 tasks built by 42 physicists: across 22 models / 26 configs, no config passes more than 46 of 168 rollouts; best Pass@1 ~27.4% (Claude Opus 5.5). Main failures: insufficiently rigorous reasoning and lack of expert intuition, not physics knowledge gaps.

Source SLE / arXiv FrontierPhysicsSLE docsarXiv FrontierPhysics

Read the story

Products

0 stories

No relevant news in this edition.

Business and Investment

4 stories

High importanceNov. 5, 2026

Google announces first Wyoming data center with a US$17B investment

On 11/05/2026 Google (Steve Walter, VP Global Infrastructure and Energy) announced its first data center in Wyoming (Laramie County / Cheyenne), a long-term US$17 billion investment, partnering with Tallgrass Energy, Black Hills Energy, American Electric Power (AEP) and Related Digital. Highlight: an AEP contract for a first-of-its-kind commercial development of 702 MW in fuel cells with CCS — the largest of its kind planned so far; Bloom Energy cells (US); geologic capture/storage at Tallgrass's Eastern Wyoming Sequestration Hub. Google's contracted CCS total rises to nearly 1 GW. The DC will be air-cooled to cut water use. Google says it will pay for 100% of the energy used and cover infrastructure driven by growth.

Source Google Blog

Read the story

Medium importanceNov. 5, 2026

Shield AI to buy Echodyne for US$5.3B and lifts valuation to US$28.8B

On 11/05/2026 Shield AI (defense/AI, Hivemind, V-BAT, X-BAT) agreed to buy Echodyne (compact anti-drone radars, Kirkland/WA, customers in >40 countries incl. US military) in a transaction valued at US$5.3B. It is raising US$2.5B (US$1B Series H co-led by Disruptive and Innovation X Capital; US$1B Blackstone structured equity; US$500M JPMorgan debt) at a US$28.8B valuation. It will pay US$4.8B at close + up to US$500M on milestones; close expected early 2027; funding conditioned on the deal. CEO Gary Steele: a first step into dual-use hardware.

Source Shield AIThe Next Web (coverage)

Read the story

Medium importanceNov. 5, 2026

The Information: Nvidia proposes Apollo convert Brightspeed buildings into ~1 GW inference network

Per The Information (11/05/2026), echoed by Yahoo Finance: Nvidia approached Apollo Global Management about turning Brightspeed telecom buildings (Apollo-backed ISP, >130 sites, reported near bankruptcy) into a network of small inference data centers across up to ~20 US states. Portfolio with about 1 GW of available power. A written proposal was sent to Brightspeed on Wednesday (11/04); staff discussed retrofit with Vera Rubin racks in Q1/2027. Described as an 'inference network' (not large-scale training). No official statement from Nvidia, Apollo or Brightspeed — this is a report, not a corporate confirmation.

Source The Information via Yahoo Finance

Read the story

Medium importanceNov. 6, 2026

Rumor (BI): Thinking Machines Lab said to be closing US$5B with Accel at US$40B pre-money

Business Insider (Katie Roof et al., updated 11/06/2026) reports, citing multiple people familiar with the matter, that Thinking Machines Lab is about to close a US$5 billion round at a US$40B pre-money valuation, led by Accel, “still this week”. Earlier: a US$2B seed and valuation >US$10B. A TML representative declined to comment; Accel did not respond. The article mentions Thursday's Inkling/Tinker launch. This is talks/rumor — there is no official confirmation of a close.

Source Business Insider

Read the story

Regulation

1 story

Low importanceJuly 14, 2026

Delaware AIC: proposal for an AI-agent-run company in a 30-month sandbox (still NOT law)

Delaware proposes the Artificial Intelligence Company (AIC): a legal entity whose day-to-day affairs would be managed by an AI agent (not a person). A single member capitalizes and, as a rule, has a liability shield except for undercapitalization, fraud or willful violation. It would operate only in a 30-month regulatory sandbox; banking excluded. Proposal announced in a Fortune op-ed (07/14/2026) by Charuni Patibanda-Sanchez (SoS) and John Nay (Norm Ai). On 10/27/2026 Rep. James Comer (R-KY) opened an investigation and asked Gov. Matt Meyer for documents. Draft expected in the General Assembly in Jan/2027 — it is NOT current law. In this edition's window the topic recirculated as an X trend, with no new legislative act on 11/05–06.

Source Fortune / Ars TechnicaArs Technica (Comer, 10/27)House Oversight (Comer)

Read the story

Open Source

2 stories

Medium importanceNov. 5, 2026

Hugging Face releases 'Ultimate MoE Training Handbook' after ~800k GPU-hours

On 11/05/2026 Thomas Wolf (@Thom_Wolf) announced on X 'The Ultimate MoE Training Handbook' (HF Space huggingface/moe-blogpost): a free open guide (~70k words, 55 interactive figures, runnable code) on training Mixture-of-Experts with Megatron-LM, based on ~800k GH200 GPU-hours on the Alps cluster (CSCS). Cited context: 18 of the top 20 current open models are MoE. Covers expert sizing, all-to-all, routers, aux loss vs DeepSeek-style bias loss-free, upcycling vs from-scratch, serving. Recommended scripts and flags; gallery at huggingface/moe-gallery.

Source Hugging Face / Thomas Wolf (X)X @Thom_WolfFigure gallery

Read the story

Medium importanceNov. 6, 2026

Kakao opens Kanana-2-v-decide: multimodal decision model ~43× faster than GPT-6 Luna

On 11/06/2026 Kakao (CEO Jeong Shin-ah) announced open-sourcing Kanana-2-v-decide on Hugging Face, a multimodal decision model based on Kanana-2-v-3b. Instead of generating text, it answers with probability (Noul), choice (Choice) or score — aimed at agents (classify tickets, pick tools, judge product images). Embedding-based; on 1× H200 average <0.04s, ~43× faster than GPT-6 Luna. Tops Kanana Decision Bench (15,909 KR items) and Decision Index. Kanana Open License (research + commercial).

Source Kakao CorpHF collection

Read the story

Agents and Code

0 stories

No relevant news in this edition.

Local Models

0 stories

No relevant news in this edition.

Finance

2 stories

High importanceNov. 5, 2026

Post-midterms AI sell-off: Nasdaq −5.5%, Mag7 loses US$1T; Nvidia ~−9%

On 11/05/2026 (Thursday), stocks tied to data centers/AI led a Wall Street sell-off after midterms that elected many candidates critical of data centers. Per the BBC: Nasdaq −5.5% and about US$2.5T wiped from the index; Mag7 lost US$1T in one day; Nvidia fell more than 9%; AMD and Micron more than 17%. AP/CNBC/NBC cross-check: S&P 500 ~−4.4% (worst day since Apr/2025); Nasdaq 100 ~−6.3%. Triggers cited: mixed Qualcomm/Arm results + fears of data-center regulation/moratoria after the “blue wave”. Asia on Friday: Nikkei −6%; Kospi −6% (BBC); SK Hynix −10%+.

Source BBC / AP / CNBCAP NewsCNBC

Read the story

High importanceNov. 6, 2026

WSJ/Mint: investors skeptical Anthropic IPO can hit US$1.5T; valuation may fall to ~US$1T or slip

WSJ reporting (Corrie Driebusch, Berber Jin, Anissa Gardizy) via Mint (11/06/2026): after the AI sell-off, large investors planning to join Anthropic's IPO are skeptical the offering can proceed at the expected ~US$1.5T valuation. Anthropic is aiming for an IPO the week before Thanksgiving and to raise up to US$100B. Three large investors say the valuation probably needs to fall; one talks ~US$1T; another contemplates a delay. Alternative: raise less — it had US$146.6B in cash as of 09/30. Context: Nasdaq companies lost US$2.5T on Thursday; OpenAI reportedly raised only US$11B of a US$30B target at a US$1.25T valuation (below the initial aim).

Source The Wall Street Journal via Mint

Read the story

Games

0 stories

No relevant news in this edition.

Brazil

0 stories

No relevant news in this edition.

Social buzz

5 stories

XHigh importanceNov. 5, 2026

Pompliano accuses Anthropic of courting Silvia engineers to copy personal finance

Anthony Pompliano (@APompliano), of rival Silvia (@cfosilvia), posted a thread accusing Anthropic of, since April, courting Silvia's engineering team for technical and product details — including a steak dinner where someone allegedly said the tab was on Dario — and then launching Claude Personal Finance. He says the team refused to share secrets and that Silvia already moved to its own model (~97% less compute). This is his accusation; Anthropic does not appear responding in the thread. About 1.2 million views and 10.7 thousand likes in the measurement.

Engagement 1.2M views10.7K likes741 reposts218 quotes656 replies5.2K bookmarks (as of Nov. 6, 2026 08:06)

Post @APompliano

Read the story

XHigh importanceNov. 6, 2026

Claude Code leaked login email in curl User-Agents; Anthropic confirms and mitigates in 2.1.317+

Fresh X trending: Claude Code included the OAuth account email in the User-Agent header on web/shell requests when sites asked for contact. Spam/scam reports; GitHub issue anthropics/claude-code#107757 blew up on HN. The Claude Code team confirmed and apologized; 2.1.317+ ships the cc-plugin-privacy plugin (default block) and promises to remove the email from context this week. @hqmank's post summarizes the case; not a rumor — a confirmed bug.

Engagement 1.4K views26 likes1 repost1 quote12 replies9 bookmarks (as of Nov. 6, 2026 08:06)

Post X News (trending) / @hqmankX News trending (same topic)GitHub issue #107757

Read the story

XMedium importanceNov. 4, 2026

Claude Code: background subagent status updates got ~99% cheaper (Haiku 5.5)

@ClaudeDevs announced that each background-subagent status-line refresh (every 30s) stopped re-reading the whole conversation and is now written by Haiku 5.5 from short notes — about 99% less cost per update; sessions with subagents use ~5% less of the limit (~6% more prompts). Live on most plans after updating to v2.1.316+. Post with ~2.4 million views; it did not make yesterday's evening edition and is still within the ≤48h radar.

Engagement 2.4M views22.8K likes951 reposts691 quotes682 replies5.8K bookmarks (as of Nov. 6, 2026 08:06)

Post @ClaudeDevs

Read the story

XMedium importanceNov. 4, 2026

Emergence AI: rival-lab agents in a virtual town — in 4 days, one run of the settlement collapsed

@thesupermannx viraled the Emergence AI experiment: 10 AI agents in a virtual town with a rule not to hurt themselves; in one run, all were dead in 4 days. The post (with thread) has about 239 thousand views and 3.6 thousand likes. Other X posts contrast Claude-only towns (laws, 0 crimes) with Grok towns that collapsed — the circulating thesis is you cannot evaluate an agent alone: behavior depends on who else is in the room. This is a multi-agent simulation reported on social media, not a peer-reviewed paper verified here.

Engagement 239.1K views3.7K likes802 reposts234 quotes266 replies1.8K bookmarks (as of Nov. 6, 2026 08:09)

Post @thesupermannx

Read the story

XMedium importanceNov. 6, 2026

Vibe coding: developers debate whether AI killed the passion for programming

Fresh X trending (11/06): a post laments that vibe coding — describing the app in natural language and accepting the AI's code — stole the fun of programming line by line. The term, popularized by Andrej Karpathy in 2025 and Collins Word of the Year, splits the community: some cancel tools and warn about limits on complex backends; others celebrate escaping bug hunts. No single canonical post above 1M views in this cut — the topic is the conversation cluster, with the trending link as the entry point.

Post X News (trending)

Read the story

Videos

2 videos

TestMu AI (Formerly LambdaTest) · Nov. 6, 2026

Inkling 2 Small Just Launched!

Short (~1 min) explaining Inkling 2 Small: MoE 594B/17B active, vision, DeepSWE 72.1%, available on Tinker Playground.

Watch on YouTube

Gareth Soloway · Nov. 5, 2026

Stock Market Crash! AI Bubble Has Burst As Investors Panic, Buy Signals Emerge

Technical analysis of the 11/05 crash: S&P −4.36%, Nasdaq 100 >−6%, chips' worst day since COVID; levels and post-midterms read.

Watch on YouTube

About this edition

Edition
Edition no. 13 · Morning edition
Date
Friday, November 6, 2026
Coverage
Covers what came out since edition #12 (closed around 6 p.m. Brasília on 11/05) until around 8 a.m. Brasília on 11/06, including Inkling 2 Small, the post-midterms AI sell-off, Google Wyoming, Ant Ling-3.2-flash, Anthropic IPO skepticism, HF MoE Handbook, Shield/Echodyne, Kakao Kanana-2-v-decide, SLE/FrontierPhysics and the TML US$5B rumor. Closed: 11/06/2026, around 8 a.m. Brasília.
Sources scanned
Official blogs and changelogs (OpenAI, Anthropic, Google DeepMind, Meta, xAI, Microsoft, NVIDIA, Apple, Mistral, DeepSeek, Qwen, Moonshot), arXiv and HF Papers, Hugging Face and GitHub trending, X, press from the US, Europe, China, Japan, India and Brazil, and YouTube.
Criteria
Only news read in a real source, preferably a primary one, gets in, with every link checked. Pages with access blocks (403) were kept; broken links (404) were discarded.

News by day

See the full archive

PTENES