Wednesday, November 4, 2026 Edition no. 10 · Evening
PTENES
Noticiário do Joãozinho Search

Artificial intelligence news from Brazil, in English

Noticiário do Joãozinho: Edition no. 10 · Nov. 4, 2026 · Evening

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. 10 · Nov. 4, 2026 · Evening

Moonshot releases open-weight Kimi K3.1 Pro: multimodal MoE of ~2.8T (104B active) and 1M-token context

On 11/04/2026 (~16:00 UTC / 13:00 BRT), Moonshot AI (@Kimi_Moonshot) announced the open-weight release of Kimi K3.1 Pro on Hugging Face (moonshotai/Kimi-K3.1-Pro). The model is a multimodal agentic MoE: about 2.8 trillion total parameters and 104 billion active, with a 1-million-token window. On Terminal-Bench 4.0 it scores 65.7% (Claude Opus 5.5 66.4%; GPT-6 Astra 58.2%); on Kimi Code Bench 3.0 Long-horizon, 61.2% vs K3’s 55.95%. On the cyber xdev-bench-hard it reaches 81.0% after data filtering (90.5% without the filter). The custom “Kimi K3.1 Pro” license is in the repo. The official blog had already promised weights by Nov 4; the X release post had about 259k views and 5,184 likes in the watchdesk measurement.

Sources

PDF (pt-BR)

Models

1 story

High importanceNov. 4, 2026

Tencent launches Hy4 with vision and computer-use; weights and system card “in the coming days”

Tencent announced Hy4 (11/04): for the first time with native visual understanding and computer-use (CLI and GUI), with specialized training on Blender and CAD. On WorkBuddy, average score 83.18 and pass rate 96.15%. In a blind evaluation with 213 finance professionals (579 tasks), Hy4 scored 3.04 — ahead of DeepSeek-V4.1-Flash-1017 (2.97), Kimi K3.1 Pro (2.68) and GLM-5.3 (2.68). Hallucination rate fell from 12.95% in the preview to 7.2%. Weights and system card come “in the coming days”; responsible open-source focus cites defensive cybersecurity and biosafety. Per TechNode, the model has 770 billion parameters; API at US$0.834 / US$2.501 per million tokens (input/output); WorkBuddy and CodeBuddy free from Nov 4–15. There is no official active-parameter count for the final Hy4 in this edition’s sources.

Source TencentTechNode (770B; API pricing; free 4–15/11)

Read the story

Research

1 story

High importanceNov. 4, 2026

Anthropic: Claude helps map the first 3D vault-particle structure in bacterial life/phages

Anthropic’s biology lab published (11/04) the first 3D structure of a vault particle found in prokaryotic cells — assembled from bacteriophage protein L05: 96 copies, about 9.5 MDa, ~54×45 nm, with about 25% more volume than mammalian vaults. Optimized purification took about 90 minutes (vs ~3 days conventionally); the first structural model came in ~34 hours via cryo-EM. Claude helped with database search, DNA constructs, isolation and microscopy processing. Vaults’ biological function remains unknown; a preprint was submitted to bioRxiv (link to update when live).

Source AnthropicOfficial X post (@AnthropicAI)

Read the story

Products

0 stories

No relevant news in this edition.

Business and Investment

1 story

High importanceNov. 4, 2026

Meta raises US$15B in bonds across 5 tranches to fund AI expansion

Meta sold US$15 billion of bonds in five parts (11/04), with yields from 5.85% to 7.53%; the book reached about US$58 billion (~3.9× the deal). The 10-year priced at 6.55%; the longest (2056) at +2.2 percentage points over Treasuries — about 0.9 pp wider than a comparable April deal. S&P cut Meta’s outlook, citing AI spending strain. In the same window, Oracle raised US$10B and Alphabet €6B; Meta is preparing its first euro bond.

Source FinancialMarkets.com

Read the story

Regulation

0 stories

No relevant news in this edition.

Open Source

0 stories

No relevant news in this edition.

Agents and Code

0 stories

No relevant news in this edition.

Local Models

0 stories

No relevant news in this edition.

Finance

1 story

High importanceNov. 3, 2026

Revolut cuts coding inference cost ~90% with self-hosted GLM-5.3 on 56×B200

At AI Engineer NY, David Abitbol (Revolut ops/infra) described the pivot: ~2,000 engineers, coding time +60% and lead time +70% after the coding-AI rollout; the default became self-hosted GLM (from 5.1 to 5.3) on 56 Nvidia B200 GPUs. Cost ~US$0.30 / million tokens — about 1/10 of Codex/Claude Code — with ~90% cut in coding inference spend. Stack: NVFP4, disaggregated prefill/decode; sub-3s TTFT at ~256k context (Abitbol: “long context is a scam”); ~200 tok/s. A router sends vision to Gemini/GPT Luna; commercial frontier stays the exception.

Source The Register

Read the story

Games

0 stories

No relevant news in this edition.

Brazil

0 stories

No relevant news in this edition.

Social buzz

4 stories

XHigh importanceNov. 4, 2026

Nebius Token Factory puts Kimi K3.1 Pro on day-0 the same day weights drop

@nebiustf (Nebius Token Factory) announced day-0 support for Kimi K3.1 Pro on the same day Moonshot released the open weights — underscoring the race among inference platforms to offer the model as soon as it ships.

Engagement 23.1K views173 likes12 reposts11 quotes12 replies12 bookmarks (as of Nov. 4, 2026 18:10)

Post @nebiustf

Read the story

XMedium importanceNov. 4, 2026

Reading the Kimi K3.1 Pro license: free up to US$20M SaaS / 100M MAU or US$20M/month; then a deal or brand credit

@guilleflorvs’s post summarizes the custom “Kimi K3.1 Pro” license: free use up to US$20 million SaaS revenue / 100 million MAU or US$20 million/month revenue; above that, a deal or brand credit is required. Waivers via Moonshot or partners — the post cites Nebius and Vercel AI Gateway with day-0.

Engagement 664 views4 likes3 replies4 bookmarks (as of Nov. 4, 2026 18:10)

Post @guilleflorvs

Read the story

XMedium importanceNov. 4, 2026

Trending on X: “weight-space programming” — frontier models editing smaller nets’ weights

X’s trending aggregator highlights buzz around “weight-space programming”: the idea of frontier models editing weights of smaller nets (e.g. Qwen3.5-0.8B) without classic retraining. Treat as social echo — do not take the X summary’s percentages as paper fact.

Post X News (trending)

Read the story

XMedium importanceNov. 4, 2026

Trending on X: miHoYo and AI plans for NPCs/agents in games + sales drop in China

Per X’s trending aggregator, a hook is circulating about miHoYo investing in AI for NPCs/agents in games alongside reports of hit-title sales drops in China. The X hook mentions “15 Billion”; without a confirmed primary press release in this edition, the figure is not treated as fact — only as aggregator echo.

Post X News (trending)

Read the story

Videos

2 videos

AI WITH Rithesh · undefined 2026, undefined

Tencent Hy4 IS HERE! INSANE Demos Tested… But My 300-Test Benchmark Tells a Different Story !

Coverage on Hy4 launch day: vision/computer-use demos and an in-house ~300-test benchmark — useful to contrast official marketing with independent measurement.

Watch on YouTube

Raidar · undefined 2026, undefined

Kimi K3.1 Pro 5x CHEAPER than Opus 5.5: a Spider-Man game for $35

Hands-on cost and coding test of Kimi K3.1 Pro versus Claude Opus 5.5 — on the same day Moonshot released the open weights.

Watch on YouTube

About this edition

Edition
Edition no. 10 · Evening edition
Date
Wednesday, November 4, 2026
Coverage
Covers what came out since edition #9 (closed around 8 a.m. Brasília on 11/04) until around 6 p.m. Brasília on 11/04, including watchdesk leads (Hy4 at 08:39 and Kimi K3.1 Pro weights at 13:34). Closed: 11/04/2026, around 6 p.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