Max reasoning should mean best results, right? 🤔 Not always. Claude Sonnet 5.5 wrote less mergeable code at max than at xhigh, one level below, while using more tokens than any model measured. 🧪 Test your settings before you trust them. 📬 deeplearning.ai/the-batch/cl… #DeepLearningAI #AIEngineering #AIAgents

Save $100 on your ticket with code X100

Save $100 off the ticket price with code X100. Limited time only.

🚀Build your AI engineering skills at AI Dev NYC: knowledge graphs, agent memory, harnesses, self-improving agents, inference and security. Plus a fireside chat with Andrew Ng and Yann LeCun on world models. 📅Nov 30–Dec 1 · Chelsea Piers, NYC ai-dev.deeplearning.ai/

Models perform worse as their context grows. In Context Engineering, built in partnership with @trychroma you'll learn to give agents the right information, in the right format, at the right time. Taught by @jeffreyhuber with guest speaker @dbreunig. Enroll for free: bit.ly/4BxLVA3

💻Learn by building with an AI coding agent, right in your browser. 👀Live preview, nothing to install. 📚Try DeepLearning.AI’s new agentic coding lab in our course, Building AI Assistants with On-Device Memory. 👉Enroll for free: deeplearning.ai/courses/buil… #AI #AIAgents #AIEngineering

🧠 Yann LeCun helped invent the convolutional neural network and shared the 2018 Turing Award with Hinton and Bengio. Now he joins us at AI Dev NYC to share his latest work at AMI Labs. 📍 NYC 🗓️ Nov 30 to Dec 1 👉 ai-dev.deeplearning.ai/ Just. Keep. Building.

🧭In this week's letter from Andrew Ng: AI improves fastest on tasks you can verify. Judgment tasks are harder, and it's easy to overgeneralize progress in math and code to the rest of human activity. Also inside: 🤖Claude Haiku 5.5 and Sonnet 5.5 🇪🇺Mistral Large 4 💻Computers for local AI 📊CausalMix 📬deeplearning.ai/the-batch/is… #DeepLearningAI #AIEngineering #MachineLearning

1/ 📰Missed a few days of AI news? Here's what happened in this week's Data Points 👇 2/ 🪶Claude Haiku 5.5: $0.10/$0.50 per million tokens for prompts up to 100K, ~90% cheaper than Haiku 4.5. Super fast and highly capable. 3/ 🎯Microsoft-Decision-1: calibrated probabilities over fixed options in one pass. Built for classification, routing, agent gating. This decision model costs $0.042 per million input tokens, and output is free. Check independent benchmarks to compare it with Jev and other decision models. 4/ 🔌Meta's Muse Wonder: Comparable with top models from Google and OpenAI, costs $4/$20 per million tokens. No full safety report yet. 5/ 📄Gemini 4 Argon: the official wide release has a 1M token output limit, up from 64K. Input and output share the window. $2/$10 intro pricing (per million tokens of input/output) until Nov 30. 6/ 🐍Python 3.15 is out, but 3.10 is end of life. Update your projects to support the new features where you can (3.11 remains a good backwards-compatible option). 7/ 📬Subscribe to our Data Points newsletter for authoritative, short-form, real-time AI news: deeplearning.ai/the-batch/ta… #DeepLearningAI #AIEngineering #Python

🧰 Picking the most relevant tools one by one leaves gaps. HYSET scores tools as a set. 📊 Recall@5: 88.6% vs. 81.4% ✅ Tasks resolved: 69.9% vs. 61.8% Full story in our weekly newsletter for AI news and research, The Batch 📬 hubs.la/Q04zXp-p0 #DeepLearningAI #AIAgents #AIEngineering

🦾 FLUX 3 Action, a video-first world-action model, ranks first on RoboLab-120’s simulations, with a success rate of 42.9%. 🎥 It was pretrained on more than 95% video. 🔧 It adapted to a new arm with about 200 demonstrations. ⚠️ It hasn’t been tested on real robot limbs at the independent RoboArena yet. Full story in our weekly AI newsletter, The Batch 📬 hubs.la/Q04zXgPK0 #DeepLearningAI #Robotics #OpenWeights

🎯 Gemini 4 Argon has a 15% hallucination rate on AA-Omniscience, compared with 51% for GPT-6 Astra. 📊 It ranks first on the Vals Index, reflecting its high performance in legal and other knowledge work. 🔐 For now, only cybersecurity defenders in Google's Fairwind program can use it. Full story in our weekly AI newsletter, The Batch 📬 hubs.la/Q04zX8YW0 #DeepLearningAI #AIEngineering #Cybersecurity

📊 GPT-6.1 Sol: 52 on the AA Intelligence Index, at $0.72 per task. GPT-6 Astra: 53, at $3.26 per task. ⏸️ OpenAI also canceled GPT-6.1 Astra’s release after the model showed deception in internal tests. Full breakdown in The Batch 📬 hubs.la/Q04zWKLz0 #DeepLearningAI #AIEngineering #AIAgents

🖥️ Agents don't just consume tokens for input and output like humans do with early chatbots. Agents use computers: searching, storing, analyzing, transacting, at a scale no person matches. In this week's Andrew's Letter in The Batch, Andrew Ng explains why that means we need much more infrastructure, and faster tools, not just faster models. 🏗️ 📬 hubs.ly/Q04zR8n20 #DeepLearningAI #AIEngineering #AIAgents

🏗️ In this week’s letter from Andrew Ng in The Batch: One of his agents makes 5,000 to 10,000 web searches a day. Agents will need far more data centers. Also inside: 🤖 GPT-6.1 Sol ⚖️ Gemini 4 Argon 🦾 FLUX 3 Action 🧰 HYSET 📬 hubs.la/Q04zPZP60 #DeepLearningAI #AIEngineering #AIAgents

A new agentic research model and harness called AREX checks its answers requirement by requirement, keeps what's verified, and researches only the gaps. 📰✅ 📊 82.5% BrowseComp, 82.0 F1 WideSearch-en 📈 Harness alone: up to +10 pts 🧠 Fine tuned 4B beat untuned 35B on 5 of 6 benchmarks hubs.la/Q04ztJKG0 #DeepLearningAI #AIAgents #LLMs

This assistant remembers where you left your keys. No network connection. No remote server.  Build one yourself in a course built in partnership with Qdrant and taught by Dylan Couzon.  👉 Enroll for free: hubs.la/Q04zGYrF0 #AI #VectorSearch #AIEngineering

Missed a few days of AI news? Here’s what happened in this week’s Data Points 🧵 🔹Google is back at the frontier. Gemini 4 Argon ties GPT-6 Astra at 53 on the Artificial Analysis Intelligence Index, at about 60% of the cost per task. 🔹Europe closes the gap. Mistral Large 4, a 1T-parameter model with open weights coming late October, is now the most intelligent model built outside the US and China. 🔹Reflection AI’s open-weight Beam claims 3–4x less compute to reason than comparable open models. 🔹OpenAI published a batch of new math results from an unreleased model, with proofs formalized in Lean. 🔹Microsoft’s MAI-Transcribe-2-Streaming ranks #1 of 38 for streaming speech-to-text accuracy. 🔹Google’s EmbeddingGemma 2 runs multimodal search on a phone; Cohere’s Embed 5 targets enterprise RAG. 🔹Meta’s Muse agent and your friends’ data, OpenAI’s opt-in text watermarks, arXiv’s new submission caps, and which AI subscriptions give the most value per dollar. Read it all 👇 hubs.la/Q04ztBjc0

Agents read more than they write, so DeepSeek shrank what they remember. Read the technical breakdown in this week's issue of The Batch. 📰🗜️ 📉 890 bytes of cache per token, 437x smaller than DeepSeek-V1 ⚡ +25% compute per output token from 4K to 1M input 🏆 Flash beats V4-Pro: 39 vs 36 on AA Index, $0.27 vs $0.67/task hubs.la/Q04ztt000 #DeepLearningAI #LLMs #AIAgents

Most voice assistants work like a relay race: speech becomes text, text goes to a model, the answer becomes speech again. Every handoff adds a pause. ⏱️ Speech-to-speech models skip the relay. Google's new Gemini 3.8 Live models listen, reason, and respond in one system. 🎯 The Extended Thinking version ranks first on Artificial Analysis' Speech to Speech Index 🗣️ The standard version ranks second in blind live conversations judged by people 💰 The standard version costs $0.84 per hour of input audio, the lowest in the index Both models also take in image and video input. Picture asking an assistant for help with whatever is on your screen, and it simply answers. 📱 Read the full story in The Batch 👉 hubs.la/Q04ztmlL0 #DeepLearningAI #VoiceAgents #AI