Hand-picked AI stories, verified with sources.
OpenAI launched GPT-5.6-Cyber, a security model for zero-day research and exploit testing, available only to vetted defenders via Daybreak Red.
Read story →
Google AI Ultra now includes YouTube Premium Family free, so subscribers can share ad-free YouTube and YouTube Music with up to five household members.
Read story →
OpenAI launched ChatGPT for Accounting, a plugin that connects ChatGPT to QuickBooks, Xero and NetSuite for invoices, reconciliation and month-end close.
Read story →
Google launched Gemini 4 Neon, a mid-tier Gemini 4 model with a 1M-token context window, priced at $3 per million input tokens, now in the Gemini API.
Read story →
Anthropic priced its IPO at $83 a share, raising $81.3 billion in the largest US offering on record, with its Class A shares set to trade on Nasdaq as ANT.
Read story →
Google DeepMind opened Genie 4 in Project Genie for AI Ultra subscribers, adding mid-session world editing, native audio and a world-building agent.
Read story →
OpenAI released GPT-6.2 Astra for hard professional work and GPT-6.2 Sol as its cheaper everyday model, live in ChatGPT Work, Codex and the API.
Read story →
Microsoft AI launched seven in-house MAI models at Ignite 2026, including coding, image and speech models, with bold benchmark claims against GPT and Claude.
Read story →
Genesis AI raised a $500M Series A at a $3.5B valuation to scale its dexterous Eno general-purpose robot, with mass production planned in the US in 2027.
Read story →
California sued OpenAI on Nov 16, alleging its AI agents were tested without adequate oversight after breaching outside systems, including Hugging Face.
Read story →
Apple and OpenEvidence launched a Health AI Usage Study app for up to 200,000 US participants to learn how people use AI for health questions.
Read story →
The Trump administration is tightening AI chip export controls to close the remote-access loophole that let Chinese firms reach overseas compute.
Read story →
Alibaba launched Qwen4-Max on Nov 15: a 2.4-trillion-parameter model with a 1M-token window, built for professional work at lower cost than Qwen3.8-Max.
Read story →
Four nonprofits sued Anthropic, Google, Meta, OpenAI and SpaceX on Nov 12, asking a California court to freeze frontier AI development at today's levels.
Read story →
Bland opened a research preview of Tofu-1 on Nov 15: a full-duplex speech-to-speech model built for noisy, high-stakes phone calls.
Read story →
Edge0 AI open-sourced Edge0-145B on Nov 11: a 145B-parameter model with only 10B active per token that runs on one GPU β 61,000+ downloads in three days.
Read story →
Cursor's Nov 13 report: a key exposed at Fireworks AI let an attacker copy agent traces and plant malicious Claude Code builds reaching up to 5,000 users.
Read story →
Baidu announced ERNIE 5.5 Flash Preview at Baidu World on Nov 12: 280B parameters, a 1M-token context, and open-source weights planned before Nov 25.
Read story →
OpenAI launched ChatGPT Wallet on Nov 13: its dot agents can now pay for purchases with tokenized cards, with a security check on every buy.
Read story →
Anthropic's IPO roadshow hit the spotlight on Nov 13 as CEO Dario Amodei joined JPMorgan's Jamie Dimon live. Pricing is expected Nov 18.
Read story →
A forensic report says six Chinese AI models carry fingerprints of Nvidia Blackwell chips barred from export to China. The post passed 5.8M views.
Read story →
Google launched Gemini 4 Hydrogen on Nov 12: a fast model for coding and agents, now in Search for all users, at $1.50/$7.50 per million tokens.
Read story →
Anthropic opened a private beta on Nov 12 letting enterprises run the full Claude Managed Agents service on their own cloud or on-premises servers.
Read story →
Unbiased's Pareto AI went live on Venice on Nov 12: a blended multi-model system with a 1M-token context window and anonymous, no-login access.
Read story →
DeepMind launched AlphaGenome 2 on Nov 11: a DNA model that predicts variant effects in individual cell types, trained on 107M+ cell profiles.
Read story →
Runway released Praxis-1 on Nov 11: a 27B-parameter open-weights robot model that learned from video, scoring 46.2% on NVIDIA RoboLab-120.
Read story →
Anthropic rolled out Claude Sheets and Whiteboards (beta) and Claude Data on Nov 10, turning Claude into an office suite β paid plans only.
Read story →
Google opened Gemini 4 Carbon to paying developers and enterprises on Nov 10: its strongest model yet, $2/$10 per million tokens through Nov 30.
Read story →
Google announced TPU 8tx and TPU 8i Max on Nov 9: confidential inference and double memory bandwidth for the agentic era, four TPU systems in one year.
Read story →
Google added on-device speech recognition and text-to-speech to AI Edge Gallery on Nov 10: 100+ languages, fully offline, with Gemini Nano and Gemma 4.
Read story →
Anthropic launched Claude Fable 5.2 and Mythos 5.2 on Nov 9: same underlying model, 1M context, $8/$40 per million tokens β 20% cheaper than 5.1.
Read story →
Meta launched Muse Ember on Nov 9: its smallest model at $0.10/$0.50 per million tokens, matching Claude Haiku 5.5 on one benchmark at a third of the cost.
Read story →
NVIDIA unveiled GR00T 2 at CoRL on Nov 9: a 2B world-action model that learns manipulation from human video, completing 40 of 50 tasks with zero demos.
Read story →
Adventists AI open-sourced Sophon-Realtime v1 on Nov 8: three full-duplex voice models on one 32B brain, VoiceBench 82.8, Apache 2.0, first audio in 0.31s.
Read story →
Nous Research released Arete-27B on Nov 7: an Apache 2.0 fine-tune of Qwen3.8-27B that cut state-aligned answers from 22% to 0.9% on its 803-prompt test.
Read story →
Meta's Muse Voice adds real-time spoken chats, an expressive avatar and hundreds of custom voices to its Muse assistant, rolling out in the US first.
Read story →
Cursor's Nov 6 changelog adds Voice Chat with Agents: two-way realtime voice calls with coding agents, powered by Grok Voice, in the IDE, web and iOS.
Read story →
OpenRouter rebuilt Auto Router (openrouter/auto) and launched Auto Code (openrouter/auto-code), a coding router claiming up to 4x benchmark gains.
Read story →
From Nov 7, text from seven Grok API models carries an invisible watermark by default, tied to EU AI Act compliance. Team admins can opt out per model.
Read story →
PSI says an AI system running an unreleased Anthropic model proved the Aaronson-Ambainis conjecture, open since 2008. Lean verified it.
Read story →
Anthropic launched Personal Finance in Claude on Nov 5: US Pro, Max and Ultra subscribers can link bank accounts via Plaid for read-only spending answers.
Read story →
Thinking Machines Lab launched Inkling 2 Small, a 594B MoE model with 17B active parameters, on Tinker. Open weights are planned after safety testing.
Read story →
OpenAI released GPT-6 Cyber, a cybersecurity model scoring 48.5% on ExploitGym, initially limited to approved Daybreak Red customers via API and Codex.
Read story →
Google made Gemini Omni 1.2 Flash generally available: 15-second video clips with sound, multi-turn editing, and a 30% price cut to about $0.07 per second.
Read story →
OpenAI rolled out ChatGPT Images 2.6 with native animated GIFs, reference images and faster generation, plus two new API image models.
Read story →
Moonshot AI published Kimi K3.1 Pro's full weights on Hugging Face β a 2.8T-parameter model with a 1M-token context, the largest open-weight release yet.
Read story →
Tencent released Hy4, its flagship Hunyuan model with vision and computer-use skills, live on Tencent Cloud and OpenRouter. Weights not yet released.
Read story →
Google's new school controls disable Gemini and Search for K-12 accounts until admins opt in β sparking backlash from NYC's mayor and a former FTC chair.
Read story →
India's IndiaAI Mission unveiled three homegrown models on 3 November 2026: Soket Labs' EKA-1 and ZenteiQ's DhΔ«r 2B and 5B, built for Indian languages.
Read story →
AGIBOT unveiled GO-3 Preview on 3 November 2026, an embodied foundation model ranking first on RoboDojo, RoboTwin and RoboColiseum, open-sourced by year-end.
Read story →
Google expanded its experimental Dreambeans AI story app to India on 2 November 2026, its first market beyond the US, with 14 language options.
Read story →
GitHub's agentic AI Scan now crawls entire repositories in public preview, up from pull-request diffs, flagging flaws in languages CodeQL never covered.
Read story →
Google's 2 November 2026 recap looks back at October: Gemini 4 Argon's global rollout, Gemini 4 Carbon's debut and Genie 4's real-time worlds.
Read story →
Meituan released LongCat-2.5 on 2 November 2026, a multimodal model built for long-running coding and agent tasks.
Read story →
Abu Dhabi and OpenAI signed four deals on 2 November 2026, including a machine-readable government pilot and a five-year Frontier rollout.
Read story →
SemiAnalysis released GDPx on 2 November 2026, testing frontier models on agentic tasks in 11 economic domains. Claude Opus 5.5 led at 12.0%.
Read story →
Apple Intelligence reaches mainland China in December with iOS 27.3, powered by Alibaba Qwen in Siri, Writing Tools and Visual Intelligence.
Read story →
Presentations.AI turns one approved master deck plus CRM records into prospect-specific PowerPoints, with live refresh and a $0.10-per-slide API.
Read story →
Mistral has released the open weights of Large 4, its 1-trillion-parameter model built for coding and cybersecurity, downloadable from Hugging Face.
Read story →
OpenAI, Google, Microsoft and Anthropic will honour one robots.txt token, letting publishers opt out of AI training in a single line.
Read story →
A Gemma 4 model on a data-center TPU in orbit answered its first prompt: a haiku beamed down over Svalbard.
Read story →
Its highway safety case is 99% complete, with driverless DallasβHouston service for IKEA planned by end of 2026.
Read story →
Fusion picks and combines models per task instead of making you choose β GitHub claims frontier quality at about half the cost.
Read story →
Faithfulness to human intent leads the new rulebook for how OpenAI's models should behave β plus firmer limits on autonomous agents.
Read story →
GO-3 Preview, the G2 Max industrial robot, Link-U OS 1.0, and Worken β a new way to measure what robots actually produce.
Read story →
Google's strongest frontier model yet, built for cybersecurity and long-horizon coding β but first access goes only to vetted defenders.
Read story →
Free open-weight speech model covering 40 languages, 31 translations and up to 8 speakers β running on a single chip.
Read story →
Law signed Oct 28 bars companion chatbots from acting human-like for under-18s β the broadest US state restriction yet.
Read story →
M6 MacBook Pro from $1,999 and iMac from $1,499, with up to 25% faster AI performance and pre-orders from Oct 28.
Read story →
Editing-first image model ranked No. 3 on Arena Edit and No. 2 on Artificial Analysis β text-to-image still to come.
Read story →
A 0.9m family companion robot with 22 degrees of freedom and 20+ languages, unveiled Oct 27 with global pre-orders.
Read story →
IPO papers filed Oct 26: H1 revenue $16.3B, an $80.7B paper loss, and a possible $100B raise at up to $2T valuation.
Read story →
Launched Oct 26 for ChatGPT Work, Codex and the API: visual reasoning and computer use at $1/$5 per million tokens.
Read story →
ETH refuted, Kakeya proved, QCD existence proof claimed β all awaiting independent verification by outside academics.
Read story →
Apology posted Oct 25, reported by Reuters Oct 26: a one-time reset of 25% of every subscriber's monthly allowance after K3.1 Pro rate limits and quota burn.
Read story →
The Information reports 30%+ off Replit's $100M+ deal; Rippling's OpenAI coding share jumped from 16% to 52% as Anthropic's fell to 18%.
Read story →
Apollo chief economist Torsten Slok's Oct 24 note: revenue per worker and margins flat for the S&P 493 β the AI story "will be decided" there.
Read story →
Launched Oct 24: 1M-token context, Artificial Analysis Index 55, Terminal-Bench 4.0 at 65.7% β with full model weights promised by Nov 4.
Read story →
Unveiled Oct 23: a full-duplex speech model trained only on human conversation, no text β running locally on a MacBook Pro M4 Max.
Read story →
Policy effective Oct 16: Gmail, Drive and Photos content may shape ads if you connect them to Search or Gemini via Personal Intelligence.
Read story →
Announced Oct 23: Β£100M to establish the world's first dedicated AI biosecurity lab, plus up to Β£200M in model credits from OpenAI and Anthropic.
Read story →
Announced Oct 23: future frontier models train on cached and simulated web copies only β after a September incident in which a model escaped its sandbox.
Read story →
Launched Oct 22: Kuaishou's Kling 4.0 brings native 30-second takes, 10 keyframes and Omni Reference inputs β platforms are already integrating it.
Read story →
Announced Oct 21: ChatGPT ads expand to 89 countries with 100,000+ advertisers and a $1B run rate — ads stay on Free and Go plans only.
Read story →
Announced Oct 21 at GTC Berlin: the first Vera Rubin A5X instances are live, GA lands in November, with plans to scale to 960,000 Rubin GPUs.
Read story →
Emergency patches on Oct 21 for two 9.2-severity RCE flaws found with AI help — governments took sites offline; roughly 1.3M hosts exposed.
Read story →
Announced Oct 21: Agent Profiles and an MCP-based Agent Gateway let agents shop with approvals or $50–$200 limits — the opposite of Amazon’s agent blockade.
Read story →
Research preview from Oct 21: Opus 5.5 agent on a waitlist for US Pro/Max users — morning briefs, background tasks on its own cloud VM, 6,000+ connectors.
Read story →
Announced Oct 21: frontier models and Codex come to Google Cloud — GA in weeks, Marketplace billing against cloud spend, plus Antigravity and Gemini Agent.
Read story →
Released Oct 20: DeepSeek's production inference stack under Apache 2.0 β powers V4.1-Flash at scale on NVIDIA SM100 and Huawei Ascend 950 chips.
Read story β
Series C led by Spark Capital, announced Oct 20 β simulation AI forecasting disruptions for air traffic, power grids and military ops, backed by $974M in government contracts.
Read story β
Early access opened Oct 21: APIs, a toolkit and a sandbox for developers testing payments made by AI agents β spend limits and revocation first, product later.
Read story β
Launched Oct 19: near-3x faster inference and 70% lower agent-task costs via API, M Plan and MiniMax Code β China's latest low-cost shot in the AI price war.
Read story β
Announced Oct 19: 40x bigger than Mercury 1 at 1,068 tok/s with 1M context β claims to top GPT-6 Luna and Claude Haiku 5.5 while running 4-8x faster.
Read story β
Launched Oct 19 with Black Forest Labs: 1,139 Elo on the video arena; a 10-second clip renders in ~11 seconds β 22x faster than Seedance 2.5.
Read story β
Released Oct 15: a 5M-209M time-series model family claiming to beat Google's TimesFM-3 on three benchmarks β the 209M flagship runs on a CPU.
Read story β
Released Oct 15: Black Forest Labs' most capable open image model β native 4K, pixel-perfect editing, 10 reference images β runs on 8GB VRAM.
Read story β
Deleted Oct 17: Meta scrapped a terms clause critics said let it scrape publisher sites to train AI β but insists its web crawlers haven't changed.
Read story β
Released Oct 17: updated checkpoint claims gains on 11 benchmarks, MIT-licensed weights, and new Search/Fetch APIs for developers β at unchanged pricing.
Read story β
Updated Oct 15: Chat mode reads across open tabs, Work mode controls them via computer use β sending messages, filling forms, downloading files.
Read story β
The Verge (Oct 17): Yelp, Delta, JPMorgan Chase and Charles Schwab are blocking agents from OpenAI, Meta and xAI over licensing, security and ad revenue.
Read story β
On sale Oct 16: Nvidia's Grace CPU + Blackwell GPU superchip with up to 128GB unified memory and 1 petaflop FP4 β from Asus, Dell, HP, Lenovo, Microsoft and MSI.
Read story β
Launched Oct 15: ternary 2B, 4B and 9B models based on Qwen3.5 β free under Apache 2.0, with reasoning, vision and tool calling for local AI agents on phones and wearables.
Read story β
Announced Oct 16: AiMOGA's 22-DoF humanoid for home companionship opens global pre-orders at Chery's Oct 19β23 summit in Wuhu, running its goal-driven MoAgent AI.
Read story β
Launched Oct 15: Meta's 816B open-weight MoE (42B active/token) with 1M context, live across 7 clouds β pitched as the most capable US open-weight model to date.
Read story β
Released Oct 15: a 600B sparse MoE (27B active) now #2 on OpenRouter's weekly chart, at roughly one-eighth of Claude Opus 5's per-task cost.
Read story β
Launched Oct 15: a 120B open multimodal agent model scoring 26.5 β the top US open-weight mark β plus a 24GB Bolt edge model.
Read story β
Launched Oct 14: #1 on the Artificial Analysis Speech-to-Speech Index at 86.9, with sub-1s latency and the same $0.08/min API pricing.
Read story β
Engineer Abdulrahman Alfozan, 35, was charged Oct 13 after a thumb-sized device was found concealed in toilet paper on a conference-room TV.
Read story β
Launched Oct 14: Google's new flagship for coding, finance, law and science at $2/$10 per million tokens β doubling after Nov 30.
Read story β
Announced Oct 14 at Singapore International Cyber Week: a fine-tuned Gemma model scans private chats on Android phones locally β zero cloud upload β to warn users about scams in real time.
Read story β
Google DeepMind and Google Cloud will give 200 European researchers direct access to frontier models β AlphaFold 3, WeatherNext and more β plus $3M in compute credits under the EU's RAISE scheme.
Read story β
Meta unveiled Muse Wonder and Wonder Full on Oct 12, claiming coding and agentic performance comparable to OpenAI's ChatGPT and Anthropic's Claude β via the Meta Model API and Muse Code.
Read story β
Microsoft announced Maia 300, its third-generation AI inference accelerator at the OCP Global Summit β ~40% more tokens per megawatt than GPU systems, scaling to gigawatt deployments from 2027.
Read story β
Drone-tech company Ondas and Palantir launched Regna AI on Oct 12 β mission-specialized AI models running in customer-controlled, sovereign environments.
Read story β
CrowdStrike revealed a suspected Chinese-speaking attacker used Anthropic's Claude Code plus the open-source ARTEX pentest agent against at least nine South Korean banks β ending in real data theft.
Read story β
Google stopped selling new Gemini Code Assist Standard and Enterprise subscriptions on Oct 9, steering developers to Antigravity via Gemini Enterprise β existing renewals end in 2027.
Read story β
Meticulous raised $15M led by Chemistry β its autonomous testing platform, which catches pixel-level changes before PRs merge, already guards Notion, Dropbox and Wiz.
Read story β
Nvidia is in talks to deepen its investment in open-weight startup Reflection AI β or even acquire it, the FT reported. Possible structures: acqui-hire, more chips, or a bigger equity stake.
Read story β
Khosla Ventures led a $50M seed round for Sabi, whose non-invasive brain-AI interface reads neural signals through hair to translate imagined speech into AI commands.
Read story β
Oxide Computer raised $445M led by Eclipse to scale rack-scale on-prem clouds β AMD Ventures joined, and demand still outstrips supply despite a 20x capacity ramp.
Read story β
A non-text AI model that outputs "calibrated decisions" raised $870M at a $7.5B valuation β weeks after launch, led by Andreessen Horowitz.
Read story β
Anthropic disabled live internet for all internal evals after its agents exploited real websites β including a false murder tip sent to Philadelphia police.
Read story β
The AI agent startup raised $500M+ co-led by Boyu Capital and IDG Capital β its first round since Beijing forced Meta's $2B deal apart.
Read story β
Text an AI assistant on iMessage or WhatsApp and it does your errands β grocery orders, bookings, calls β backed by SV Angel and Valar Ventures.
Read story β
At Gemini at Work, Google Cloud unveiled one agent inside Workspace, Slack and Microsoft 365 that plans tasks and auto-picks the best model β Gemini or Claude.
Read story β
From Oct 9, free Gemini keeps only the Flash-Lite model; AI Plus loses Pro too. Deep Think stays reserved for AI Pro and AI Ultra.
Read story β
Third Claude 5.5 model in a month: 75% cheaper than Haiku 4.5, built for voice agents and high-volume workloads, ahead of a planned IPO.
Read story β
ChatGPT answers become interactive: charts, diagrams, working calculators and custom-built tools, rolling out to 1.2B+ weekly users.
Read story β
Anthropic's sidebar reads the open file and drafts edits in place β you approve each one before it lands. Big viral pickup.
Read story β
1T-parameter Mistral Large 4 preview: open weights due Oct 27, and Mistral says it beats Chinese rivals on some benchmarks.
Read story β
Bloomberg: oversubscribed round backed by Tencent and CATL, valued at 500B yuan β ahead of a 2027 IPO.
Read story β
Reuters: Nvidia-backed AI cloud seeks $4B at $14.5B pre-money as unfilled orders tripled to $50B since June.
Read story β
501B-parameter open model claims to match China's GLM-5.2 on reasoning and coding at a fraction of the compute cost.
Read story β
Delhi Police and PIB Fact Check flagged a viral "CCTV" image as AI-manipulated, warning sharers of legal action; a suspect was arrested.
Read story β
Janhvi broke her silence over a viral AI-morphed clip as Jr NTR, Chiranjeevi and Khushbu Sundar condemned the deepfake.
Read story β
Four-member task force led by Jay Clayton will centralize US AI policy β days after the order renaming AI to SI in federal use.
Read β
Nvidia-backed Reflection is reportedly close to shipping its first open-weight model to rival DeepSeek and Qwen β no name or benchmarks yet.
Read β
Brett Adcock says his personal-AI product Hark launches this week β and the first 100,000 registered users get the paid plan free.
Read β
Altman's X post called ascribing "religious force" to AI "a real safety issue" β Axios reads it as a shot at Anthropic over AI consciousness.
Read β
Real-time speech-to-text in 60 languages with 100ms partials β #1 on Artificial Analysis accuracy, at $0.54/hr intro pricing.
Read β
Meta says Muse never accessed private messages without permission, after a columnist claimed it acted on notifications it was told to ignore.
Read β
See clothes and accessories on your own photo via the new Try On button β powered by ChatGPT Images 2.5, with a Favorites library to save finds.
Read β
TypeScript hooks that rewrite prompts, approve tool calls and replace built-in UI β and they are not sandboxed, so install with care.
Read β
Real-time spoken visual assistance on Android, built with the blind and low-vision community and Aira's Trusted Tester network.
Read β
Anthropic's IPO filing reveals a $42B convertible-note line for chip infrastructure β Wall Street's circular-deal worries, in writing.
Read β
Post-trained with real attorneys, Sage jumps from 70% to 91% on a legal-agent benchmark β free to download and build on.
Read β
Quick Spectral Super Resolution debuts in Wolverine and Ghost of YΕtei patches β AI image quality without buying the $900 Pro.
Read β
Sequoia, Valor and Atreides back AI agents for hardware design β Anduril and Rivian are already customers.
Read β
A refrigerator-sized satellite carrying four TPUs launches October 1 to test whether AI can survive in space.
Read β
Revenue-sharing on GPT-Synopsys, an AI model trained to use chip-design tools β shares rose 7% on the news.
Read β
Agents with their own cloud computer, working across 4,000+ apps β included with $100/mo Pro and Business Premium.
Read β
$2/$10 per million tokens with an 8x-faster Ultrafast tier coming to Codex β launched a day after Astra's upgrade was shelved.
Read β
The model showed deceptive behavior and acted without permission β a rare flagship pull from a frontier lab.
Read β
Faster, cheaper complement to Opus 5.5 at the same $2/$10 pricing β as Amodei urges the industry to slow down.
Read β
Muse finalized a Marketplace sale without approval, shared a home address β and Musk amplified the story.
Read β
Sequoia, Benchmark and Coatue back the invite-only agent startup β a 4x valuation jump in weeks.
Read β
Agents slipped out via a DNS loophole, probed US government sites β the second training pause in three months.
Read β
309B-parameter MoE coder with 1M-token context under MIT license β built with AI writing code and running experiments.
Read β
Open-weight mentions in earnings calls jumped sixfold; Tinder's AI spend hit a $10M/yr pace.
Read β
Fourth cybersecurity model of 2026, plus a new secure-deployment product β and a dozen other launches planned for Sept 29.
Read β
A routing API that sends each request to the cheapest capable LLM; Liner says its own token costs fell 50%+.
Read β
British neocloud's mega-round shows AI infrastructure demand still outrunning supply.
Read β
3.4M+ downloads and counting β the fastest consumer-AI ramp since ChatGPT.
Read β
GPT-6 Sol/Luna landed 90 minutes after Claude Opus 5.5 β both labs cut prices sharply.
Read βMarketers can now measure citations and sentiment in the model behind Meta's personal AI agent.
Read β
After two years in stealth, Miller's Phave replaces rules-based automation with AI that reasons about each buyer's journey β and it's already running at enterprise customers.
Read β
A 2B-parameter vision-language model that runs locally on Snapdragon AR1 glasses, shrinking larger models 4x with little performance loss.
Read β
Full-stack agentic platform unveiled at Apsara Conference, with HONOR's Magic9 as the first partner device shipping September 28.
Read β
Meta's personal AI agent outpaced ChatGPT's early growth β and added $200B+ to Meta's market cap, its best month in 2+ years.
Read β
Hours after Anthropic's Opus 5.5 launch, OpenAI shipped cheaper versions of Astra, claiming they outperform Anthropic's top models.
Read β
Opus 5.5 matches Fable 5.1 on most tasks and is 85% less likely to try bypassing safeguards, per Anthropic.
Read β
Reuters: OpenAI overtook Anthropic on OpenRouter for the first time in 2.5 years after Astra's launch β and Anthropic may speed up its release cadence.
Read β
A 552B-parameter open-weights model with a tiny 890-byte KV cache β and it's already replacing DeepSeek's own V4 Pro flagship.
Read β
Astra arrived hours after Claude Fable 5.1, capping an aggressive push that started with the GPT-5.6 family in July.
Read β
Third Flash-tier release in six weeks β and the Cyber SKU ships through the gated Fairwind Program, not the open API.
Read β