2026-10-30 · FridayPublished daily at 08:00
AI Daily
Mistral Large 4 (Le Chonk) arrives on Hugging Face: an open-weight, multimodal MoE model with 1T parameters

Hugging Face published a launch post for Mistral Large 4 (Le Chonk), a multilingual, multimodal MoE model with around 1T total parameters, 49B active parameters, and a 1M-token context window.
- 1
- 2OpenAI Explains How It Detects Misaligned Behavior in Frontier ModelsOpenAI: Website updates
- 3
01Model Releases/Updates
Mistral Large 4 (Le Chonk) arrives on Hugging Face: an open-weight, multimodal MoE model with 1T parameters
Hugging Face published a launch post for Mistral Large 4 (Le Chonk), a multilingual, multimodal MoE model with around 1T total parameters, 49B active parameters, and a 1M-token context window.
Original textPika Post-Trains on MiniMax H3 and Open-Sources the Pika H3 Flex Video Generation Model
Pika Labs has post-trained MiniMax H3 to create and open-source Pika H3 Flex, reducing the number of sampling steps from 50 to 6, greatly shortening generation time while improving quality.
Original text02Product Releases/Updates
Codex announces that usage limits for all Pro and Business Premium tiers will double starting today
Tibo announced that starting today, usage limits will double for the Pro 100/200/500 tiers and all Business Premium tiers. The Pro 200 limit will not be reduced, and Work and Codex usage will remain unchanged. Expanded usage limits for dots will continue to be available and will still be calculated separately from default usage. Further improvements to models and efficiency will also increase the amount of usage available in practice.
Original textFactory Releases Droid 1.0, an Enterprise-Grade Agent Harness Built Around Programmatic Tool Calling
Factory has released Droid 1.0, a model-agnostic agent harness centered on programmatic tool calling (PTC), in which models write JavaScript programs to execute hundreds of tool calls at once in a sandbox.
Original textOpenRouter rebuilds its default routing algorithm: TTFT down 16.9%, 429 errors down 67%
OpenRouter announced a rebuilt default provider routing algorithm. Instead of sorting by price alone, it uses Routing Cost to select providers based on effective price, TTFT, throughput, error rate, and output quality.
Original text03Industry News
The New York Times says Chinese intelligence operatives targeted Nvidia CEO Jensen Huang
According to The New York Times, U.S. intelligence agencies believe that an executive close to Nvidia CEO Jensen Huang provided intelligence to China. The person is Alice Huang (黄乐) of Singaporean data center operator Megaspeed.
Original textReddit Announces a Shift to a Login-First Experience in Response to Falling Traffic from AI Search
In its Q3 earnings report, Reddit announced a shift to a login-first experience to protect its data from AI scraping and improve ad monetization. In October, search referrals fell by an average of more than 1 million per week. Q3 global daily active users totaled 130.3 million, while daily active users in the U.S. fell to 50.7 million, down 2% year over year. Q3 revenue grew 53% year over year to $895 million, with net income of $283 million. The shift will first be tested outside the U.S.; there is no global launch timeline yet.
Original textGoogle’s Q3 search growth falls short of expectations as investors worry AI search monetization could weigh on its core business
Google’s search revenue grew about 12% in Q3, below analysts’ expected 15%, and its market value briefly fell by as much as 10%. Executive Philipp Schindler acknowledged that strong growth in AI Mode queries had an impact because of their lower monetization rate. The company also said annual growth in search queries was the highest in five years and that it plans to expand its chatbot ad program in the U.S. and international markets.
Original text404 Media investigates Proton’s internal AI transition dispute: CEO Andy Yen’s aggressive AI push fuels a culture of fear
404 Media interviewed nine current and former Proton employees and reviewed internal documents to report on the internal conflict sparked by CEO Andy Yen’s aggressive push to transition to AI. In October, AI wrote 90% of the company’s code, mostly using Anthropic’s Claude. Yen said this would be the last year the company hired engineers and announced a round of interviews to assess employees’ AI usage, which could determine whether they keep their jobs.
Original text04Paper research
OpenAI Explains How It Detects Misaligned Behavior in Frontier Models
OpenAI published a detailed post explaining its multilayer monitoring system for misaligned behavior in frontier models. The four-layer architecture includes activation probes, model-based review, investigation agents, and automated investigations. It monitors all RL runs and evaluations for GPT‑6 Astra and later models, as well as internal Codex traffic.
Original textTrakkr study: tracing the provenance paths behind 1285 AI citation statistics
Trakkr Research asked ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews 200 questions requiring numerical answers and traced the citation chains for 1285 statistics. Only 40% led to research consistent with the claim, 8% led to research but with altered numbers or context, 40% could not be verified, and 11% had no traceable source.
Original text05Tips and perspectives
OpenRouter tutorial: Extracting tables from PDFs with LLMs
OpenRouter published a tutorial on sending PDFs as files, together with a strict JSON schema, to models that can read files, so tables are returned as typed rows. The field descriptions specify cleanup rules for items such as parenthesized losses, thousands separators, and footnote markers.
Original text06News Briefs
- The Verge examines Anthropic's S-1: roughly $850 billion in compute commitments bet on recursive self-improvementThe Verge:AI
- OpenAI updates Model Spec, reorganizing model behavior guidelines around five core valuesOpenAI: Website updates
- OpenMOSS releases open-source image generation model MOSS-Image, focused on long-text rendering and editable layoutsOpenMOSS / Fudan NLP / Shanghai Innovation Institute / MOSI
- Addy Osmani on why you might be overprompting ClaudeAddy Osmani
- Wiz analyzes seven key signatures of agent attacks and emerging trendsWiz:Research
- Gemini adds direct generation, preview, and editing of documents, PDFs, presentations, and spreadsheetsGemini
- Tencent Hunyuan releases the Hy-Nomas technical report and open-sources Hy-Nomas-BenchTencent Hunyuan
- Figma Explains How It Trained a Specialized Canvas-Editing ModelFigma AI / Agent / Make
- Claude Enterprise officially launches intelligent reports, in-app research, and shared team budgetsClaude:Blog