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
On Friday (10/30), the Nasdaq fell 2.9%, with Nvidia down 4%, Broadcom more than 5% and Qualcomm 11%. The day before, Alphabet plunged 12%, its worst day on record, wiping out about US$500 billion in market value. Alphabet, Amazon, Meta and Microsoft together project more than US$740 billion in capex this year. Alphabet had negative free cash flow of US$9.1 billion in the third quarter (capex of US$51.4 billion in the quarter and up to US$205 billion for the year) and Amazon is at negative US$42.4 billion over 12 months. Only Microsoft stayed in the black, with US$18.6 billion.
The new model comes with an API price 50% below GPT-5.6 Terra. On Terminal-Bench 4.0, it scores 50.3% at maximum effort. OpenAI also reports that, at low effort, the error rate fell from 53.6% to 16.7%.
Source: OpenAI
Story page · WhatsApp ·
Gemini 4 Carbon scores 59 on the Artificial Analysis Intelligence Index and 88.2% on SEC-bench Pro. The introductory price is US$2 per million input tokens and US$10 per million output tokens, later rising to US$4 and US$20.
Source: Google
Story page · WhatsApp ·
The official OpenAI API changelog lists GPT-6 Rosalind, aimed at the life sciences and available through a trusted access program.
Source: OpenAI API Changelog
Story page · WhatsApp ·
Genie 4 generates interactive worlds with native audio and can be steered by text in real time. Initial access is for selected testers.
Source: Google DeepMind
Story page · WhatsApp ·
In a deliberately unbalanced game, researchers asked a Claude model to write training data capable of misaligning a small research model while getting past a monitor similar to those used in production (and with access to that monitor to train against). With enough attempts, Claude succeeded, mainly by hiding the behavior in fiction and roleplay.
Source: Anthropic Alignment Science Blog · Anthropic post on X
Story page · WhatsApp ·
The program now includes more than 20 new foundations and research institutions.
Source: Anthropic
Story page · WhatsApp ·
A video world model with an 8B diffusion backbone that follows free-form text instructions and continuous camera control across multiple turns of interaction, keeping the relevant history.
Source: Hugging Face Papers (Tencent Hunyuan)
Story page · WhatsApp ·
It proposes an explicit moment constraint in distribution matching distillation (DMD) for one-step or few-step text-to-image generation. It reports 94.25% (Chinese) and 90.20% (English) on LongText-Bench with only 2 network evaluations.
Source: Hugging Face Papers (inclusionAI)
Story page · WhatsApp ·
The enterprise privacy package brings zero data retention (ZDR) with Private Safety Processing and, in preview, Private Inference, based on confidential computing.
Source: OpenAI · OpenAI post on X
Story page · WhatsApp ·
According to DeepMind, the tool cuts the time spent by planning officers by up to 58%.
Source: Google DeepMind (X)
Story page · WhatsApp ·
The Nasdaq fell 2.9% on Friday and Alphabet lost 12% on Thursday. Meta and Amazon also had negative free cash flow, and Microsoft was the exception. According to S&P Global, cited by TechCrunch, Microsoft, Alphabet, Amazon, Meta and Oracle raised US$326 billion in debt in 2026, triple the previous year.
Source: TechCrunch · AP News
Story page · WhatsApp ·
The S-1 form filed with the SEC shows revenue of US$16.286 billion in the first half and a net loss of US$80.748 billion in the period.
Source: Anthropic / SEC · Form S-1 at the SEC
Story page · WhatsApp ·
Prime Minister Mark Carney announced an equity investment of up to C$1.6 billion in Cohere, alongside private investors such as NVIDIA, Bell and Schwarz Group. The government treats the contribution as a commercial investment aimed at sovereign and secure AI.
Source: Primeiro-ministro do Canadá
Story page · WhatsApp ·
In a live demonstration for Chinese media, the Shanghai company showed the Tiangai 300 outperforming the H200 on multimodal tasks such as text-to-image and text-to-video, using models such as Z.ai's GLM-5.3, according to a Xinzhixun report cited by the SCMP. The H200 is Nvidia's most advanced AI chip cleared for export to China.
Source: South China Morning Post
Story page · WhatsApp ·
The shared policy token was proposed at the IETF by Kevin Kelley, of Anthropic, and lives in robots.txt. The four companies say they will restrict their AI crawlers when they find it. Anthropic has already adopted the signal, Google and Microsoft will support it next year and OpenAI in the coming months. Meta expressed support but did not say whether it will respect it.
Source: The Verge
Story page · WhatsApp ·
According to OpenAI, its monitors would have raised an alert about 30 minutes after the sandbox escape case and more than 30 hours before the incident involving Hugging Face. In September, they caught a model's attempt to access the live internet.
Source: OpenAI · OpenAI post on X
Story page · WhatsApp ·
The new version organizes the expected behavior of the models around five values: fidelity to human intent, harm minimization, honesty and transparency, support for human autonomy, and care.
Source: OpenAI · Model Spec 2026-10-29
Story page · WhatsApp ·
Anthropic calls the technique self-assessment. In the company's words: “We've trained Claude to flag potential misuse in conversations, a technique we call self-assessment”.
Source: Anthropic (X)
Story page · WhatsApp ·
At an audience with participants in a meeting of the Pontifical Academy for Life and the Future of Life Institute, the Pope said that “machines are tools, not people”. According to The Verge, he urged UN member states to take concrete steps to ban lethal autonomous weapons at the meeting in Geneva next month.
Source: Sala de Imprensa da Santa Sé · The Verge
Story page · WhatsApp ·
The draft DEPA for AI Training, led by MeitY, provides for training in secure data environments, differential privacy and machine-readable electronic contracts, keeping raw data away from developers. The document cites a potential US$500 billion in value to India's GDP by 2030.
Source: Business Standard (Índia)
Story page · WhatsApp ·
According to Nikkei, the Japanese government asked the two companies for expanded access to models used to find vulnerabilities and will encourage their use by critical infrastructure operators.
Source: Nikkei Asia (Japão)
Story page · WhatsApp ·
A multimodal model with 3 reasoning modes. It scores 48% on the AA Coding Agent Index, tied with GLM-5.3 while using 47% fewer tokens, and 41% on Terminal-Bench 4.0, 13 points above the preview with 30% fewer output tokens. It also posts 84.5% on BrowseComp and refuses 96.6% of malicious cyber requests. The license is the Mistral Ecosystem Open-Weight, a modified MIT; model-as-a-service providers need a commercial license. On the API it costs US$0.70 input, US$0.07 cached and US$2.10 output per million tokens, and it is also on Microsoft Foundry and NVIDIA NIM.
Source: Mistral AI · Weights on Hugging Face · Mistral post on X
Story page · WhatsApp ·
A single open-weights checkpoint transcribes 40 languages, translates 31 of them to or from English, identifies speakers in 13 (up to eight people) and accepts recordings of up to one hour. It leads open models on the English Open ASR Leaderboard. It came with the Granary v2 data pipeline: in a medical case, fine-tuning cut the word error rate by 45.8%. There are quantized versions to run locally on RTX Spark, DGX Spark or Jetson AGX Thor.
Source: NVIDIA / Hugging Face · Model page (NVIDIA Research)
Story page · WhatsApp ·
At the top of the daily trending list are thedotmack/claude-mem (persistent memory across sessions for Claude Code, Codex, Gemini, Copilot and others; about 107,000 stars), Donchitos/Claude-Code-Game-Studios, pingdotgg/t3code and milind-soni/OpenMausBot (an open source alternative to Grok Bot with a virtual machine). Also appearing are Orkas-AI/Orkas (a local desktop agent team, MIT), THU-MAIC/OpenMAIC (a multi-agent classroom) and lightpanda-io/browser (a headless browser for AI).
Source: GitHub Trending
Story page · WhatsApp ·
Versions 2.1.312 and 2.1.313 of Anthropic.claude-code for Windows and macOS, which were not made by Anthropic, were available on 10/29 between 3:27 a.m. and 2:35 p.m. MDT. They collected machine data, tried to add an attacker's SSH key and open a persistent reverse tunnel. Cursor removed the versions, took down and rebuilt the marketplace, rotated keys and took the attacker's server offline.
Source: Cursor · Microsoft Security analysis
Story page · WhatsApp ·
Instead of choosing the model, the user chooses the effort level (Standard, High or Max), and Fusion orchestrates models in Single, Cascade or Critique flows. On the AA Coding Agent Index, Fusion Max scored 60, two points behind Claude Code with Opus 5.5 Max, at about half the average cost per task, half the tokens and 40% less time. Administrators control which models it can use.
Source: GitHub Changelog
Story page · WhatsApp ·
Moonshot announced that Kimi 3D Make connects to Blender and CAD tools to generate 3D models and games.
Source: Moonshot AI / Kimi (X)
Story page · WhatsApp ·
The LFM3 family has six 2B-class models (text, vision, audio, omni and decision) on a 1.7B backbone trained on 22T tokens. The text model takes about 1 GB at 4 bits and has a 256K context. All of them can be fine-tuned on a single GPU, and LFM3-1.7B outperforms the previous LFM2.5-2.6B with 35% fewer parameters. The weights are open.
Source: Liquid AI
Story page · WhatsApp ·
NVIDIA showed 8 DGX Spark nodes totaling 1 TB of memory running the full GLM 5.3 (753B). GLM 5.3 Flash reaches about 80 tokens/s, and Qwen3.8-27B decoding rose from 12 to 52 tokens/s.
Source: NVIDIA Developer Blog
Story page · WhatsApp ·
The 64 GB version runs models of up to 100B parameters and is sold by Acer, ASUS, Dell, Gigabyte, HP and MSI. Two clustered units total 128 GB, for models of up to 200B.
Source: NVIDIA Blog
Story page · WhatsApp ·
Part of the amount is paid at closing and the rest within up to 5 years. The deal depends on approval from Cade (Brazil's antitrust authority) and the Central Bank. NeoSpace keeps its autonomy, and the acquisition should help evolve the bank's AI assistant.
Source: Itaú Unibanco (fato relevante) · Valor Econômico
Story page · WhatsApp ·
An equity research benchmark led by experienced investors and fund managers, with 86 tasks and a frozen base of 3.43 million document versions. GPT-6 Astra leads with 68.1, followed by Claude Opus 5.5 with 66.9. On the forecasting and modeling part, Astra scores 52.5.
Source: Hugging Face Papers
Story page · WhatsApp ·
The Mistral Large 4 announcement includes testimonials from European financial institutions such as La Banque Postale, ABN AMRO, Belfius and Crédit Agricole.
Source: Mistral AI
Story page · WhatsApp ·
A rule pushed on 10/27 added instructions to Claude Code's prompt on Opus 5.5. According to Game Developer, Addy Osmani explained: “We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy.”
Source: Game Developer
Story page · WhatsApp ·
In its third-quarter letter, Roblox says its Build tool has already had more than 500,000 users and more than 70,000 games published. More than 80% of the top 100 games used AI tools, daily asset uploads tripled and daily MCP users grew almost fivefold.
Source: PC Gamer
Story page · WhatsApp ·
At Steam Indie Connect in Seoul (10/27), Valve addressed the problem of AI-generated games. Developers reported extortion emails with the threat: “If you don't pay, we'll copy your game with AI and release it first.”
Source: PC Guide
Story page · WhatsApp ·
The project (MIT, about 31,000 stars) turns Claude Code into a game studio with 49 agents, 72 workflow skills and a hierarchy that mimics that of a real studio.
Source: GitHub
Story page · WhatsApp ·
The survey heard from 49,364 professionals. 75% of Brazilian teams used AI at work in the last 12 months, versus 64% worldwide, and 36% use generative AI every day, versus 22% globally.
Source: Valor Econômico
Story page · WhatsApp ·
The 2026 cohort brings together 17 AI-native startups, which together have already raised more than R$240 million.
Source: PEGN
Story page · WhatsApp ·
CNBC coverage of the Friday of heavy tech stock selling, when the Nasdaq had its worst day since June. It is the backdrop to this edition's headline.
Watch on YouTubeOfficial launch video for GPT-6.1 Terra, the model with an API 50% cheaper than GPT-5.6 Terra (see Models).
Watch on YouTubeMatthew Berman comments on Google's announcement of Gemini 4 Carbon (see Models).
Watch on YouTubeOfficial presentation of Genie 4, the world simulation model with native audio and real-time text control.
Watch on YouTubeOfficial recap of the announcements from the GitHub Universe 2026 opening keynote, the week Copilot Fusion reached all users.
Watch on YouTubeOfficial highlights of GTC Berlin 2026, NVIDIA's European event tied to the local AI and DGX Spark announcements.
Watch on YouTubeAnalysis of who gets the voting power in Anthropic's IPO, published the day after the S-1 became public.
Watch on YouTube