Note
New models
New AI models and updates from the labs.
87 stories. Back to Home.
November 2026
- 19 Google launches cheaper Gemini 4 Neon that matches top rivals on coding, law and science — A new lower-cost Google AI model is competitive with pricier leaders on hard tasks, and you can start using it today.
- 18 OpenAI cuts prices 20% on GPT-6.2 Astra, adds Sol model for everyday use — Two new OpenAI models ship with lower costs per task, but coding skills stay flat and hallucinations remain high.
- 17 OpenAI launches GPT-6.2 Astra and Sol — cheaper, faster AI models for work and daily use — Two new models from OpenAI promise better results for a fraction of the cost, making expensive AI tools more affordable for businesses and everyday users.
- 16 Google adds 66 regional voices to its Gemini AI assistant — Gemini can now speak in local accents from the UK, India, Brazil and more, making conversations feel more personal.
- 16 Alibaba launches Qwen4-Max, an AI that builds games and 3D scenes and reads whole books at once — The new model handles huge documents, runs complex workflows faster and cheaper, and helps build 3D scenes and games.
- 14 MiniMax H3, Alibaba Wan 3.0 lead 2026's best AI video models — with an open-weight surprise — A new ranking shows which AI can turn your words or photos into video, how much they cost per second, and why one open model is now the base for most big names.
- 14 Perplexity launches Decider v1.3, a model that classifies text and images faster and cheaper than chatbots — It lets apps instantly judge whether content is safe, decide which task needs a bigger AI, or find specific information — without writing out long replies.
- 13 Apple TV 4K gets Siri AI and a pricier jump — what you need to know before buying — Apple's newest streaming box adds the same smarter Siri as iPhones, but it costs twice what the last one did.
- 13 New Gemini 4 Hydrogen AI model joins Google's family — Google launched its cheapest Gemini 4 model, now available through a common open tool developers use daily.
- 12 Google launches Gemini 4 Hydrogen, a fast cheap AI model for everyone — A cheaper, faster Google model works now for free in the Gemini app and Google Search.
- 11 Google launches Gemini Decision 1 — an AI that scores answers without writing them out — It lets apps and developers get fast yes-or-no answers across text, images, audio, video and PDFs, priced like a utility bill instead of per word.
- 11 Meta launches Muse Ember, a cheap AI model built for repetitive agent tasks — A new small AI from Meta that does routine coding and workflow steps for about a third of the price of rivals.
- 10 Meta's Muse Ember matches Claude Haiku at a third of the cost — Meta's smallest AI model does routine agent tasks for about a third of what rivals charge.
- 10 Video Rebirth shrinks MiniMax-H3 to 13.7B so it runs faster on consumer PCs — A smaller version of a Chinese AI that makes video with sound can now run on your own graphics card instead of needing huge data-centre machines.
- 10 ByteDance says its new Seedream 5.5 model can edit one image 100 times without breaking it — A new AI image tool that lets designers keep making changes over and over — useful for anyone who has ever lost quality after a few edits.
- 10 Anthropic upgrades Claude Fable and Mythos to version 5.2 with faster speeds and lower prices — Claude Fable 5.2 and Mythos 5.2 are now available — faster, cheaper, and stronger at coding and hard reasoning tasks than before.
- 10 Meta launches Muse Ember, its cheapest small AI model for everyday agent work — Meta's new low-cost AI model aims to do routine tasks like tool calls and coding faster than rivals for less money.
- 10 Apple charges $50 for extra AI usage in its creator apps — After your monthly AI feature quota runs out in apps like Logic Pro, you can buy another pack — here’s how it works.
- 03 Netflix uses AI to help you stop scrolling and start watching — A new feature guesses what mood you're in and suggests shows or movies so you spend less time browsing and more time viewing.
- 02 Nace AI opens Drex Max v1, a 27B decision model that runs on your own Mac — Companies can now make routing, classification and policy-check calls locally on Apple Silicon instead of paying for cloud APIs.
- 01 Edge0 releases small voice-cloning model that speaks 23 languages for free — A lightweight open model lets you clone a voice from a short sample — now in many more languages.
- 01 Block unveils Dessa, a voice AI that learns only by listening to humans talk — A new AI from Block talks and listens at the same time — and it was trained only on human speech, not on text.
October 2026
- 31 Samaya Sterling a tiny AI finance agent beats a bigger one at a fraction of the cost — A small post-trained model does financial research better than a large frontier model, for 11 times less money.
- 29 OpenAI brings Oracle business apps to ChatGPT so workers can run company tasks from one chat — Companies using Oracle software can now ask ChatGPT to do finance, procurement, and planning work inside their approved business systems.
- 28 Google DeepMind’s Genie 4 makes interactive worlds with sound and live talk — The new AI can generate realistic, talking virtual environments in real time, turning any idea into a living scene you can shape with words.
- 27 Mirendil launches Fern 0.1, an AI model that can read 10 million tokens at once — A new coding AI can hold a whole software project in memory at once — useful for developers, but still far behind top models at actual coding tasks.
- 27 Baidu releases Unlimited OCR 2, an open model that reads hundreds of pages in one go — It turns full books and long reports into structured text, so anyone with a GPU can process documents that usually trip up AI tools.
- 26 OpenAI releases GPT-6.1 Terra, a cheaper AI model that can use your computer — It costs half as much as the last version and can handle tasks like reading invoices, editing videos, and filling out forms — all cheaper than ever.
- 26 Microsoft launches Decision-1, a fast AI model that scores answers instead of writing them — The new model helps software make quick decisions — like routing a support ticket or checking if content is safe — at a fraction of the speed and cost of regular chatbots.
- 26 Hugging Face's Carbon-X reads DNA like a language, answers in English — An open model lets anyone chat with raw genetic code — at a fraction of the cost of top paid tools.
- 23 Kimi K3.1 Pro, a Chinese AI model anyone will be able to download, nearly matches Claude Opus 5.5 — Moonshot AI's 2.8-trillion-parameter model scores 55 on Artificial Analysis, three points behind Claude Opus 5.5, costs half as much, and its full weights — the files that let anyone run it themselves — arrive by November 4.
- 23 Kimi K3.1 ranks well for intelligence and sits on the cost-efficiency line — The model can think through hard problems and delivers solid smarts for its price, but it writes a lot and is slow.
- 23 Kimi's new AI model approaches top US competitors in 3D design and matches them in coding — Kimi K3.1 Pro lets you build full games, 3D models, and financial analyses — now available to use today.
- 23 Recursive Releases Flint, a 40B Open-Weight Local Coding Model for Agents — Flint is Recursive Superintelligence's first model: a fast local coding agent running on consumer hardware.
- 23 NVIDIA Kumo Relational beats supervised ML on relational data without training — A new pre-trained foundation model predicts outcomes directly on enterprise databases, requiring no feature engineering or task-specific training.
- 23 Inception Releases Mercury 3 Diffusion LLM at 1,068 Tokens per Second — A new AI model generates full answers in parallel, claiming top speed and strong coding scores against GPT-6 Luna and Claude Haiku 5.5.
- 22 Inception Labs releases Mercury 3, a diffusion-based AI model — The new model claims to match top competitors while running 4 to 8 times faster, proving that diffusion AI can scale to complex tasks.
- 22 YC startup builds cheap AI model that catches silent agent failures (2026-10-22) — Laminar’s flow-1 model detects hidden errors in AI agent workflows at a fraction of the cost of top commercial models.
- 22 YC startup builds cheap AI model that catches silent agent failures — Laminar’s flow-1 model detects hidden errors in AI agent workflows at a fraction of the cost of top commercial models.
- 22 vLLM Hyper-Optimizes Kimi K3 on NVIDIA GB300 for Blazing-FAST Inference (2026-10-22) — New optimizations push Kimi K3 decode to over 840 tokens per second, setting a new bar for AI inference speed.
- 22 Meta launches Muse Spark Open, an open-weight model built to run on Macs (2026-10-22) — The new 816-billion-parameter model can run locally on Apple computers, but you will need serious hardware to make it work.
- 22 Meta launches Muse Spark Open, an open-weight model built to run on Macs — The new 816-billion-parameter model can run locally on Apple computers, but you will need serious hardware to make it work.
- 22 Microsoft trains MAI-Base-2, its most efficient frontier foundation model yet — The new MAI-Base-2 model uses far fewer computing resources than rival open-weight models while delivering state-of-the-art results.
- 22 StepFun Releases Open Weights for Its 600B Step 5 Preview Model (2026-10-22) — The Chinese AI lab has made its large Mixture-of-Experts model available for self-hosting, targeting developers building coding agents and finance tools.
- 22 StepFun Releases Open Weights for Its 600B Step 5 Preview Model — The Chinese AI lab has made its large Mixture-of-Experts model available for self-hosting, targeting developers building coding agents and finance tools.
- 22 Gemini 4 Argon Tries Its Hand at StarCraft Bot Coding (2026-10-22) — Google's new model ranks third in a benchmark where AI systems write bots for the classic real-time strategy game.
- 22 Gemini 4 Argon Tries Its Hand at StarCraft Bot Coding — Google's new model ranks third in a benchmark where AI systems write bots for the classic real-time strategy game.
- 22 A New AI Model Learns to Predict Many Possible Futures at Once — Researchers introduce a system that doesn't just predict one future, but maps out entire branching paths — and it leads to better decision-making.
- 21 Mistral Large 4 is an open-weight multimodal model with 1.05 trillion parameters — The French company Mistral has released its latest large language model, featuring a mixture-of-experts architecture and vision capabilities.
- 20 Mistral launches Voxtral 3, a voice AI system for 50+ languages — The new model family brings speech transcription, natural voice generation, and a real-time agent SDK together in one package.
- 19 doubleAI introduces Maxwell, a model that runs inside code to speed it up — A new kind of AI model doesn't write programs — it lives inside them and makes them faster on the spot.
- 19 MiniMax launches M3.1 Flash a cheap agentic AI model with a 1 million token context — The new MiniMax M3.1 Flash model matches top AI agents at less than half the price of its older sibling.
- 19 Mistral releases Timestral, a new family of time-series forecasting models — The French AI lab says its largest model tops every major benchmark while running on a standard CPU.
- 19 DeepSeek releases V4.1 Flash 1017 Max with 1M token context and reasoning ability — The new open-weight model scores top marks in intelligence and speed benchmarks from Artificial Analysis.
- 18 Google Unveils Gemini 4 Argon with 1 Million Token Output — The new frontier model can generate a million tokens in one go, but trails rivals on real coding tasks.
- 17 Google Antigravity Speeds Up Search for New Antimicrobial Peptides — Google DeepMind's agentic platform combines AI models and interactive workflows to find potential new antibiotics in minutes instead of months.
- 16 Liquid AI Unveils d1-ultra Decision Model Ranking 1 on AI Benchmark — The new model reads text and images to answer questions with probabilities instead of generating text, cutting costs for developers.
- 16 NVIDIA Releases Open 120B Nemotron 3.5 Super VL Model — The open-source model combines text, image, video, and coding into one agent-ready system that runs on a single GPU.
- 15 OpenAI Brings GPT-Rosalind to Academic Research Labs Through New Free Plan — The Rosalind Academic Plan offers eligible labs up to 30 free seats to access OpenAI's life sciences model for up to 12 months.
- 15 FLUX.3 Image Klein Arrives in Hugging Face Diffusers — The 6-billion-parameter image-to-image model from Black Forest Labs is now available through the popular open-source library.
- 14 T0 A 256M open-weights model redefines time-series forecasting — The Forecasting Company released T0, a small but powerful foundation model that produces probabilistic forecasts for everything from electricity demand to drug reimbursements.
- 14 Simon Willison Updates llm-gemini to Support New Gemini 4 Argon Model — The developer behind the popular llm plugin adds Google's latest low-cost model, which promises strong performance at a fraction of rival prices — at least until December.
- 14 New Plugin Lets You Use Meta’s Muse Wonder AI Models — Simon Willison updated his llm-meta-ai tool to work with Meta’s latest top-tier reasoning models.
- 13 Google rolls out Gemini 4 Argon with heavy safety defenses — The latest Gemini model gets robust protections against misuse and misalignment as it opens to the public.
- 13 Liquid AI releases open-weight model built for fast on-device agents — The new 24-billion-parameter reasoning model runs 1.7 times faster than rivals on edge hardware.
- 13 Meta Unveils Muse Wonder and Muse Wonder Full as Its Most Capable AI Models Yet — Meta's newest models claim frontier-level performance on agent tasks and will soon power its Muse AI assistant.
- 13 Google's Gemini 4 Argon brings deep physical reasoning to real robots — The robotics-focused Gemini 4 Argon model scores 87% on a hard 3D reasoning benchmark and controls Boston Dynamics Spot in real-world manipulation tasks.
- 13 Google's Gemini 4 Argon targets the hardest problems first — The new frontier model from Google DeepMind is already helping engineers and scientists tackle months of work in weeks.
- 13 NVIDIA launches compact AI model for smarter game NPCs — A new lightweight language model lets players talk to in-game characters that respond in real time, running locally on consumer hardware.
- 13 Meta Launches First Frontier AI Model, Muse Wonder — The tech giant enters the top-tier AI race with two new models built for complex tasks and personal agents.
- 12 Meta's Muse Wonder Full breaks into AI's top tier — Meta's new frontier models close the gap on leaders, with coding leaps and tougher pricing.
- 12 Meta's Muse Wonder Brings It Back to the AI Frontier — Meta's newest models rival the best from OpenAI and Anthropic, marking a major comeback after years of falling behind.
- 11 Microsoft Unveils Microsoft-Decision-1, a Decision AI Model Built on Alibaba's Qwen3.5 — A new kind of AI that picks from a list instead of writing text could make agents faster and cheaper.
- 11 OrcaRouter Releases OrcaCyber Zero 1.5 With 1M Context for Cybersecurity Research — A gated AI model with strong vulnerability-finding benchmarks launches at a fraction of the price of rival frontier tools.
- 10 Microsoft launches Decision-1 to compete in fast-growing AI decision model market — The new model aims to deliver fast, accurate decisions for routing and classification tasks, adding to a crowded field of specialized AI tools.
- 10 How Qwen-Image-2.1-Turbo Generates Images in Just 8 Steps — Alibaba's accelerated checkpoint cuts denoising by 5x using the same 7B architecture, prefix KV caching, and classifier-free guidance disabled.
- 10 Google’s Unreleased Gemini 4 Carbon Matches Anthropic’s Opus 5.5 on Coding — An internal testing variant is matching the best coding model from a rival, and it hasn’t even launched yet.
- 09 Compressed 27B Model Beats Full Version at Tool Calling in Under 8 GB — A heavily compressed version of Qwen3.8-27B runs on everyday hardware and outperforms the original on agent skills.
- 08 Google DeepMind Launches EmbeddingGemma 2 for On-Device Multimodal Search — The new 740M-parameter model can search text, code, images, video and audio on your phone or laptop without sending data to the cloud.
- 08 Google Launches Agentic AI for Gemini, Starting with Business Customers — Google's new unified agent can plan, execute, and complete work across business systems — starting with enterprise clients before a consumer rollout.
- 08 JetBrains Releases Mellum2.1 A Fast Open-Source Coding Model — A 12-billion-parameter model that uses real software environments for training can now be run on your own hardware.
- 08 Perplexity Releases Two Multimodal Embedding Models for Edge and Cloud — The 9B model scores 92.4% on agentic PDF QA, while the 0.6B variant runs on laptops for fast, cheap queries.
- 07 OpenAI Rolls Out GPT-6 With Interactive UI to All ChatGPT Users — The latest model brings visual, clickable responses to every tier of ChatGPT — starting with paid users on October 7 and free users the next day.
- 07 Mistral Rolls Out 'Le Chonk' Europe's Challenge to US and China AI Dominance — France's Mistral AI claims its new trillion-parameter open model is the strongest outside China, giving businesses a safer alternative to US proprietary systems.
- 07 Reflection AI Launches Beam a 501B Open-Weight Model Built for Coding Agents — A new 501-billion-parameter open model aims to beat larger competitors on efficiency for coding and agentic tasks.
- 07 Reka Unveils Rho-1 A Single AI Model That Generates Video and Controls Robots — The new 19-billion-parameter model replaces multi-model pipelines by understanding, creating, and reasoning across text, images, video, and robotic actions all at once.
- 07 Mistral AI Unveils Mistral Large 4, a 1.05-Trillion Parameter Multimodal Model — Mistral AI's new flagship model, nicknamed 'Le Chonk,' combines massive scale with a 1M token context window and strong cybersecurity performance.