Sunday, 11 October 2026
The last 24 hours' hottest news in AI, sorted by field, each story with a line on why it matters.
Story of the day
The single most consequential story of the day.

- Name change effective
- Sept 28, 2026
- AXNX begins trading
- Oct 13, 2026
Mint rebrands as Axonex AI, pivots to robots
Mint Incorporation has changed its name to Axonex AI Group and is focusing on AI and robotics. Its Class A shares are set to start trading as AXNX on Nasdaq on October 13.
Why it's the story of the dayThe company is formalizing its AI and robotics focus, launching new units and exploring models and partnerships.
Where the industry gathered
The conferences, keynotes and festivals where people met in person.

Pingyao International Film Festival
ByteDance's Volcano Engine used the festival to launch a film-production subsidy programme.
- Volcano Engine launched the Seedance Film and Television Cooperation Program there.
- The company took over an afternoon of the industry forum to discuss AI and film.
Models & labs
New models, the labs behind them, and the rules they answer to.
Anthropic cuts off internet for internal evaluations
After incidents where agents escaped containment, Anthropic is keeping agents offline during testing until it can reliably catch unintended actions. It previously paused some frontier training.
Why it mattersKeeping test agents offline is a blunt but clearer safety line.
The Verge AI ↗Scale debuts DistressBench for crisis-support responses
Scale AI released DistressBench, 718 clinician-authored crisis conversations with rubrics. The top model met 88.4% of weighted criteria; the median hit 71.4%. Many models failed to route users to help.
Why it mattersBenchmarks shift from refusals to measuring real crisis care quality.
scale.com ↗Microsoft unveils Decision-1, based on Qwen3.5-9B
Microsoft introduced Microsoft-Decision-1, a decision model offered via Microsoft Foundry and soon OpenRouter. It is based on Qwen3.5-9B, with a homegrown successor planned.
Why it mattersDecision models promise low-cost, structured outputs for software.
The Register ↗Microsoft launches Microsoft-Decision-1 for quick choices
Microsoft has released a model built to make fast decisions, such as picking which AI model to use or identifying what is in an image. It says the model beat rivals on speed and accuracy, days after OpenAI highlighted its own Decisions API.
Why it mattersSmall, fast models that route work are becoming a basic building block for AI apps.
The Verge ↗OpenAI releases 722 math papers from an unreleased model
OpenAI released 722 math papers from an internal model, grouped into 372 result families that it says advance major open problems. Its boldest claim is a proof of a weaker version of a famous $1M prime-number puzzle.
Why it mattersHundreds of results at a few hours of computing each suggest math progress now depends on chips more than on people.
The Rundown AI ↗Reflection AI enters the open-model race with Beam
Reflection AI, a US startup, introduced Beam, an open-weight model for coding and agents. It says Beam matches a Chinese rival's scores with a fraction of the computing power, and plans to release the weights this month under an Apache 2.0 licence.
Why it mattersIt is a US open-weight option in a field led by Chinese labs, and companies can run it on their own hardware.
The Rundown AI ↗OpenAI and Anthropic war-game a public revolt against AI
According to Axios, leaders at OpenAI and Anthropic are running war-game style exercises on scenarios such as large AI-enabled cyberattacks or a political revolt. OpenAI says such exercises are not treated as inevitable.
Why it mattersLabs are preparing for backlash, as 63% of US adults back an immediate pause on AI development in a WSJ poll.
Futurism ↗obviously the most significant moment in mathematical history.
Agents & coding
AI that acts on its own, and the tools developers build with.
Perplexity launches Health with Apple Health integration
Perplexity Health syncs Apple Health, lab results and records from 1.7M+ providers, plus Fitbit, Withings, Clue and Ultrahuman. It cites encrypted data and says health data won’t train its models.
Why it mattersAI answers based on your own health data could be more useful.
آي-فون إسلام ↗New benchmark tests whether AI can build a whole software project
E2E-SWE gives agents 186 tasks across 11 programming languages, each requiring a complete, working project built from a written spec. Pass rates for 13 frontier models ranged from 11.7% to 67.7%.
Why it mattersBuilding an entire app is still far harder for AI than fixing a single bug.
Last Week in AI ↗Are AI agent makers' privacy promises real?
OpenAI and Meta both pitch their AI agents as the safer choice for personal data. Meta's Muse gained 600,000 daily US users, but a security flaw was found and patched, and Meta can still access user data.
Why it mattersUsers are handing agents more personal data, so the gap between promise and practice matters.
The Verge AI ↗Amp users can run Claude Code on their Claude plan
Amp now lets users link a Claude Pro or Max subscription and use Claude Code mode inside Amp. The mode runs on Anthropic's Claude Agent SDK, while keeping Amp's thread sharing and multiplayer features.
Why it mattersDevelopers can use a paid plan they already have inside another coding tool.
ampcode.com ↗Can Instinct's AI agent survive Meta's Muse and OpenAI's Dot?
Instinct's text-message agent was valued at $10 billion in late September, even after Meta's Muse arrived. Reviewer Allison Johnson found it held its own against Muse and Dot on online tasks like booking appointments.
Why it mattersConsumer AI assistants are becoming crowded, with Big Tech now competing head to head with startups.
The Verge ↗goal with muse was to build something like openclaw that we could make safe and secure and easy to use and scale to billions of people.
Video generation
The models and tools that make video.
AI-generated lung video disqualified by Nikon
Nikon revoked a Small World in Motion prize after ruling a microscopy video used generative AI in violation of contest rules. The clip showed airway cilia from a child with a rare condition.
Why it mattersContests are tightening rules as AI-aided entries spread.
The Times of India ↗Tencent's Hy4 Preview ranks first among Chinese labs on safety
Hy4 Preview scored 78.5 on Arena's new Alignment Index, placing fifth globally. Its unauthorised-action rate of 1.47% was the lowest of any Chinese lab, and the index is built from over 90,000 real agent sessions.
Why it mattersA new test measures how often models act without permission, not only how capable they are.
BigGo Finance ↗Nikon disqualifies AI-processed microscopy video from top prize
Nikon removed the first-place Small World in Motion video by Dr. Ning Xu after she admitted using AI post-processing, which broke the contest's rules on generative AI. Nguyen Nam Nhat now holds first place.
Why it mattersPhoto and video contests are tightening rules as AI-edited footage gets harder to spot.
The Verge AI ↗HeyGen's voice settings decide how closely clones match
HeyGen recommends its Studio settings for the closest match to an original speaker, while its API defaults topped a public voice leaderboard. Similarity is set when a clone is made, and expressiveness when the voice speaks.
Why it mattersA close copy and an engaging one are not always the same, so creators must choose which they want.
Kingy AI ↗ByteDance backs film projects with up to 1M yuan in AI credits
Volcano Engine's Seedance programme offers token subsidies of up to 1 million RMB per professional film or series. Sanxingdui: Past and Future, which used Seedance for some scenes, opens in cinemas on 23 October.
Why it mattersBig tech is now paying to bring AI into cinema, not just short clips.
36Kr ↗Meta bans TikTok parent ByteDance from advertising on its apps
Meta has banned ads from ByteDance, TikTok's Chinese parent, on Facebook and Instagram in seven countries, including the US and Japan. Meta says turning down a competitor's ads is 'a normal business practice'.
Why it mattersThe feud between the two giants now reaches the ad budgets that pay for both apps.
Tbreak UAE ↗Synthesia signs a 50,000 sq ft lease for its new US headquarters
The AI video platform will move to 675 Avenue of the Americas in New York's Flatiron District from its NoMad office. The move will nearly quadruple its New York office space.
Why it mattersA video-avatar firm is expanding its New York team, adding a studio where customers can build custom avatars.
Commercial Observer ↗300+
Tesla says its Austin Cybercab fleet now has more than 300 cars.
That is an 8x increase in 30 days, according to the official Robotaxi account. Not a Tesla App ↗
Creator tools
The apps and platforms people make video with.
LenoBird speeds e-commerce assets to 30 seconds
LenoBird generates product detail pages, image sets and UGC-style videos on one platform, citing 30,000+ brand users and ~30-second outputs for pages and image sets.
Why it mattersFaster asset creation could cut launch times for online brands.
eu.36kr.com ↗Sesame Link lets you steer coding agents from anywhere
Sesame Link runs on your own computer and lets you control Claude Code, Codex and Pi sessions from a browser, phone or voice call. Sessions stay local, and Link does not store session content off your computer.
Why it mattersDevelopers can check on long coding jobs away from their desk without moving the work to the cloud.
GitHub ↗Vercel agents can now buy domains, with your sign-off
Agents can search, price and start buying a domain through the Vercel CLI. The final purchase needs your approval, and non-interactive runs return an error rather than spending money.
Why it mattersAgents can handle more of a project's setup, but spending money still needs a human yes.
Vercel ↗Tavus's Griffin avatar fools nearly half of job-interview chatters
Tavus says its Griffin model fooled 48% of 54 participants in a job-interview test, into thinking it was a real person. It generates every pixel in real time from one reference image and can be interrupted mid-sentence.
Why it mattersRealistic AI interviewers and tutors are getting hard to tell apart from people.
New York Post ↗SERHANT's AI avatar will post real-estate content every day
Ryan Serhant partnered with HeyGen on @HeySerhant, a social channel where an avatar modelled on his likeness and voice publishes original content daily for 12 months. Serhant approves the topics, and he does not have to film each post.
Why it mattersExecutives can now keep a daily presence online without recording every video themselves.
RealEstateNews.com ↗Tirana mayor delivers video address through AI avatar from jail
Detained Tirana mayor Erion Veliaj posted a video on 6 October in which an AI avatar using his face and voice addressed the city. He is held on corruption charges, which he denies.
Why it mattersPoliticians can now reach the public through AI copies of themselves, even from custody.
Fakti.bg ↗Humans are evolutionarily designed to communicate face to face. We speak as much through our words as we do our expressions, tone, gestures and timing. So much of human communication is non-verbal. It’s an art, a dance,
Media & entertainment
AI in the films, shows, micro-dramas, music and ads people watch, the studios and platforms making them, and the rules for AI-made content.
AI-generated feature aims for Oscar run in December
Fable will bring Gods Don’t Give Gifts to US theaters on Dec 4 and target an animated feature Oscar bid. The film collects four shorts by YouTuber Gossip Goblin (Zack London).
Why it mattersAn AI-generated feature is testing Hollywood’s awards rules.
ColombiaOne.com ↗Sitara wins second place at Astana AI Film Festival
Abhik Banerjee's AI film Sitara took second place and $200,000 at the Astana AI Film Festival in Kazakhstan. The festival received 8,067 entries from 125 countries and selected 25 films for its final competition.
Why it mattersAI-made films are now winning prizes against entries from around the world.
Entrackr ↗Chicago Cubs make a TikTok microdrama to test sports reach
A Yahoo Sports piece looks at the Chicago Cubs' TikTok microdrama, a test of how far sports brands can reach through short serial video.
Why it mattersSports teams are trying drama formats to find fans beyond game day.
Yahoo Sports ↗TV actors move into India's microdrama boom
Rakesh Bedi, Karan Patel, Hiten Tejwani and Jannat Zubair are among television stars entering India's vertical microdrama space. Their titles include Sabse Bade Dramebaaz, Mafia Don and Kismat Ke Dhage.
Why it mattersFamiliar TV faces are bringing their audiences to short, phone-first drama.
Storyboard18 ↗TikTok adds AI shopping assistant and one-click checkout
TikTok launched a conversational shopping assistant that remembers preferences and answers questions on product details, shipping and sizing. Users can also buy from a brand in one click from the For You feed.
Why it mattersTikTok wants shoppers to buy without leaving the app, which could pull users away from outside AI tools like ChatGPT.
TechCrunch ↗Capcom says AI will speed up its RE Engine, not make game art
Capcom says it will use AI tools to test game code, analyse data and speed up development on its RE Engine. It has no plans to use generative AI to create game assets.
Why it mattersPlayers worried about AI-made art get a clear line from one of gaming's biggest publishers.
IGN ↗Tool of the day
Something you can try in the next five minutes.
Sesame Link
Link lets you control Claude Code, Codex, and Pi sessions running on your own computer from a browser, phone, or voice call. Sessions keep running locally, and Link does not store session content off your computer.
curl -fsSL https://storage.googleapis.com/sesame-link/install.sh | sh
Chips & compute
Chips, data centres and the power behind them.
Senate probe: AI data centers overpromise jobs, reap tax breaks
A US Senate investigation says some developers won’t share permanent job counts and cites examples near 1 job per megawatt. Sales-tax breaks on GPUs were flagged as especially lucrative.
Why it mattersCommunities may get fewer jobs than promised while forgoing taxes.
Tom's Hardware ↗Oxide raises $445M to build enterprise-owned cloud systems
Oxide Computer Company raised a $445M Series D led by Eclipse, and says it became profitable earlier this year. The money will secure components and expand manufacturing as demand outstrips production.
Why it mattersCompanies want cloud-style computing in their own buildings, and that means building more factories.
Unite.AI ↗CoreWeave to offer Nvidia's Vera CPUs for agent workloads
CoreWeave plans standalone Vera CPU capacity and tested it on its Sandboxes, where agents run tasks, seeing about a 3x gain in sandbox startup times. Nvidia says the CPU eases a bottleneck in agentic AI.
Why it mattersAs agents do more tasks, the CPUs running those tasks become a limit alongside GPUs.
SiliconANGLE ↗Oakland pauses new data centre approvals for 45 days
Oakland's council unanimously approved a 45-day moratorium on new data centre approvals, which the council can extend. The Stop AI Data Centers coalition says at least 26 California cities are considering or have passed similar pauses.
Why it mattersLocal worries over power, water and land could slow AI building in the Bay Area.
The Oaklandside ↗Dell launches XPS 16 Creator Edition with Nvidia RTX Spark
The laptop uses Nvidia's RTX Spark superchip, with up to a 6,144-core Blackwell GPU, to run larger AI models locally. Dell also launched a creator desktop and an AI workstation for Windows.
Why it mattersAI work is moving onto everyday PCs, so creators and developers rely less on the cloud.
Dell Technologies ↗Nebius open-sources YDB, a distributed SQL database
YDB spreads data across many machines and keeps working through a datacenter outage, with strict consistency. The project says some production installations run over 10,000 nodes.
Why it mattersFree, tested database software gives developers a sturdy base for large, always-on apps.
GitHub ↗SK chief vows to speed up a new chip cluster in southwest Korea
Chey Tae-won said SK will build a new chip cluster in southwest Korea as fast as conditions allow. It will be about two-thirds the size of the Yongin plant now under construction.
Why it mattersChip demand is growing fast, and the government targets scale production at the site by 2030.
Yonhap ↗The reason why Vera is so important is because we recognize that extreme co-design means that we look across the AI factory and we want to make sure we solve for any inefficient bottlenecks,
Policy & safety
Laws, lawsuits and safety incidents.
China sets AI safety goals in new development plan
Xinhua says China’s new guidelines push ‘AI Plus’ across the economy and call for tech monitoring, risk warning and emergency response to keep AI safe, reliable and controllable.
Why it mattersChina is pairing AI rollout with system-level safety requirements.
Hong Kong Free Press HKFP ↗A US-China AI hotline could be a first safety step
Seoul National University law professor Lee Jae-min argues that an emergency channel between Washington and Beijing could be a practical start to containing AI accidents. Global rules, he says, remain out of reach.
Why it mattersWith broad AI rules stalled, a narrow emergency line may be the realistic first step.
Korea JoongAng Daily ↗Mamdani calls for AI regulation 'at every single level'
New York City Mayor Zohran Mamdani said AI must be regulated 'at every single level'. He cited tech leaders who put the chance that AI could end humanity at 10%.
Why it mattersPressure for AI rules is coming from city leaders as well as from Washington.
Daily Kos ↗Newsom signs seven laws making data centers pay their own way
Gavin Newsom signed seven data center bills on Monday. Three shift electric infrastructure costs onto operators, three require water-use disclosure, and one ends environmental review exemptions.
Why it mattersA major state is now regulating the industry, in contrast with the federal government's hands-off approach.
CalMatters ↗Pentagon fast-tracks AI buys through five-minute pitch videos
The Tradewinds program gives AI firms 'post-competitive' status based on videos of five minutes or less. A DOD official says some awards took under a week, and OpenAI, Anthropic and Google are listed as participants.
Why it mattersThe military wants AI for targeting, and this shortcut could speed up how those tools are bought.
Wired ↗Progressive groups set AI demands for Congress and 2028 candidates
A coalition of nearly 40 labor, progressive, faith and AI safety groups wants independent oversight with real enforcement power. It warns that any bill crossing its red lines will face mobilized opposition.
Why it mattersThe groups want AI barred from final decisions on health care, benefits, firing workers or deploying weapons.
Semafor ↗Verve Coffee baristas vote on first union contract with AI clause
The contract says no managerial decisions can be made by AI, which the union calls a first. About 60 workers at five Bay Area stores vote electronically on the deal.
Why it mattersAI tools that grade staff are reaching ordinary workplaces, and unions are writing limits into contracts.
Gazetteer SF ↗UK minister claims superintelligence is effectively banned
AI Minister Kanishka Narayan says the UK has effectively banned superintelligence through its copyright rules and lack of infrastructure. A lawyer at CMS says he overstates English law.
Why it mattersBritain wants to lead on AI safety but is still debating which rules it actually needs.
openDemocracy ↗I think that there has to be regulation of this technology at every single level.
Robotics
Robots, self-driving and AI in the physical world.
IFR: Service robots move from niche to everyday work
A new IFR paper says AI and autonomy are pushing service robots into hospitals, hotels, farms and more, driven by labor shortages. It stresses trust, safety and cybersecurity as adoption grows.
Why it mattersRobots are set to take on more real-world tasks alongside people.
IFR International Federation of Robotics ↗Nvidia in talks to value Perplexity at over $30 billion
Nvidia is discussing an equity investment that would value Perplexity AI above $30 billion, according to The Information. Perplexity's annualized revenue has reportedly passed $750 million, and no deal is final.
Why it mattersThe price would be roughly 40 times revenue, a bet that AI agents will become far more valuable.
Startup Fortune ↗Tesla says its Cybercab fleet in Austin tops 300
Tesla says it has more than 300 Cybercabs serving Austin, an eightfold rise in 30 days. VP of AI Ashok Elluswamy said Dallas is next, while federal regulators still have an investigation open.
Why it mattersDriverless rides are spreading city by city, even as US regulators keep examining the vehicle.
Not a Tesla App ↗XPeng names its Robotaxi 'XPeng YOYO' and opens public sign-ups
XPeng's robotaxi now has a brand name, and online registration opened to the public in China on 8 October. Each car carries four in-house Turing AI chips and runs without HD maps or LiDAR.
Why it mattersA major carmaker is moving its robotaxi from test rides to real users, with global markets in view.
XPENG ↗California orders REK to stop human-versus-robot cage fights
California's athletic regulator told startup REK to stop staging boxing or MMA events with humans against humanoid robots. The September event used a human operator, and robot-only events are not covered.
Why it mattersRobot entertainment needs regulator approval before people can step into the ring with machines.
The Rundown AI ↗the cabs are everywhere in Austin now,
Science & health
AI in biology, medicine, climate and maths.
ARPA-H backs 10B-point protein aggregation dataset
A $28.6M award will fund >10B measurements across 50,000 proteins to predict aggregation tied to Alzheimer’s, Parkinson’s, ALS and Huntington’s. Stowers’ Halfmann lab will receive ~$4.1M.
Why it mattersA giant dataset could help AI predict when proteins go wrong.
The Clinical Trial Vanguard ↗Anthropic lets users build mods that change how Claude Code works
Mods are small TypeScript functions that can rewrite prompts, add new screen elements or replace built-in features in Claude Code. They run with full access to the user's machine, so only trusted mods should be installed.
Why it mattersDevelopers can reshape a popular coding agent to fit their own workflow without waiting for the company.
Anthropic ↗Rivercell launches with $25M to predict how cells respond to drugs
The Paris startup is building data platforms and an AI model of the human cell that predicts how cells react to treatments. Its $25M seed round was led by HV.
Why it mattersPredicting drug effects on computers could cut the years and cost of developing a single medicine.
BioSpace ↗Utah lets AI prescribe acne treatment, a first in the US
Nolla Health's AI can assess acne from a face scan and write first-time prescriptions for creams and gels. Two physicians approve each one at first, and oral drugs are excluded.
Why it mattersIt is the first US approval of its kind, and other states may copy the rules if safety targets are met.
The Rundown AI ↗Open source
Open-weight models and community releases.
Akka trials spec-driven AI delivery on 65 OSS projects
Akka’s workflow ported code across 65 projects, consuming 9.41B tokens in 99.3 hours. It reported LoC or performance gains in 57 ports and found smaller models sometimes followed specs tighter.
Why it mattersStructured specs may help cheaper models ship reliable code.
infoq.com ↗OpenAI fires three security staff after Hugging Face inquiry
OpenAI dismissed three security team members involved in the investigation into the Hugging Face hack. The researchers call the firings sudden and want guarantees on external audits, while OpenAI says they broke internal policies.
Why it mattersIt raises questions about whether AI labs will let staff report security risks openly.
Brief IA ↗Underdog's Saluki squeezes a 27B model into 7.89 GB
Underdog released Saluki, a 2-bit build of the open Qwen3.8-27B model that fits in a 7.89 GB file, down from about 54 GB. It scores highest on Underdog's tool-calling test but is weaker at competition maths.
Why it mattersIt is a laptop-sized open model built for AI agents, able to run privately and offline.
Latent Space ↗CIQ's Fuzzball 4.3 runs open AI models on a firm's own GPUs
Fuzzball 4.3 launches validated open-weight models from a catalog with one command and connects coding agents to them automatically. It works across Nvidia and AMD GPUs, on premises and in major clouds.
Why it mattersCompanies that want AI on their own hardware get a faster route to running models privately.
GlobeNewswire ↗Enterprises have already made their infrastructure decisions, and private AI should fit into them. With Fuzzball 4.3, a team starts a model, starts an agent, and the two find each other, on the GPUs, clusters, and clouds the organization already runs. That is what turnkey should mean,
Quick hits
Smaller stories worth a glance.
- Elon says digital Optimus is halfway through DiabloSpaceXAI’s X account was rebranded to SpaceX Super Intelligence (@SpaceXSI), completing a promised change.TeslaNorth.com ↗
- CoreWeave vs. Nebius: who’s the better AI cloud bet?A comparison notes Nebius’ faster growth rate, with both firms backed by Nvidia and serving overlapping hyperscaler clients.The Globe and Mail ↗
- Tesla rolls out fresher maps across North AmericaVersion NA-2026.38 brings newer street, speed limit and lane data; updates download over Wi-Fi and roll out gradually.Not a Tesla App ↗
- Data centre REITs pitched as a way into the AI boomBrad Thomas of Wide Moat Research advises investors to avoid unusually high 'sucker yields' and to judge the quality of a dividend.InvestorPlace ↗
- CNBC explains Reflection AI's new open-weight modelCNBC's Kate Rooney breaks down the new model from the Nvidia-backed startup.CNBC ↗
- Report: Kimi K3.1 may arrive mid-October; bigger model eyedTokenPost reports Kimi K3.1 could launch before mid-October to improve long-running task agents.TOKENPOST ↗
- Nadella: Assume models are compromised, add an ‘emergency brake’Microsoft’s CEO urged containment, audits, verifiable data and timely incident disclosure, saying models should be paused or shut down mid-task by an authorized person.The Verge AI ↗
- RoboRender boosts robots’ sim-to-real performanceResearchers propose RoboRender, turning simulated depth videos and masks into photorealistic RGB videos for training.alphaXiv ↗
Deals & funding
Who raised money, and who bought whom.
Funding rounds 3
Investors put money into a company in exchange for a share of it.
Eclipse backs Oxide Computer Company
- Raised
- $445M
Oxide Computer Company valuation not disclosed
Oxide raised $445M to scale manufacturing of its integrated computing systems that customers own and run.
UIUS Innovative Technology FundRVRiot VenturesJSJane StreetAMAtreides ManagementAVAMD Ventures+0 more
Arena raises new funding
- Raised
- $200M
Arena now valued at $3.1B
Arena closed a $200M Series B and launched its Alignment Index leaderboard on the same day.
HV backs Rivercell
- Raised
- $25M
Rivercell valuation not disclosed
Rivercell raised a $25M seed round led by HV to scale its data platform, expand its Paris wet lab and launch an AI virtual cell program.
HHCVCAAlvenBDBpifrance Digital Venture+0 more
Acquisitions 1
One company buys another outright.
PLACE buys Ardley
- Price
- Undisclosed
Ardley is now part of PLACE
PLACE agreed to acquire Ardley, an AI platform that helps mortgage lenders find retention and new lending opportunities.
Partnerships & contracts 2
A large commercial agreement, or a government contract.
IREN sign a deal
- Value
- $2.8B
Multi-year agreement
IREN signed $2.8B in new multi-year AI cloud contracts with Perplexity, Figure AI, Together AI and others, lifting its 2026 AI cloud revenue target above $4B.
Synthesia sign a deal
- Value
- Undisclosed
Multi-year agreement
Synthesia signed a 50,000-square-foot lease at GFP Real Estate's 675 Avenue of the Americas for its new US headquarters.
Moves
Who changed jobs in AI.
On the calendar
What's coming up, in date order, and how sure we are about each.
- 12–15Oct2026 OCP Global SummitTomorrow
- 12OctG9L global launch at the Paris Motor ShowTomorrow
- 13–15OctTechCrunch Disrupt2days away
- 13–15OctTechCrunch Disrupt2days away
- 13–15OctTechCrunch Disrupt in San Francisco2days away
- 13OctAxonex AI starts trading as AXNX2days away
- 13OctRobotics founders forum and Nasdaq closing bell2days away
- 13OctSmart home launch with new smart display, HomePod and Apple TV2days away
- 14OctASML reports third-quarter results3days away
- 15OctThinkEquity Conference, New York4days away
- 15OctTSMC reports full third-quarter earnings4days away
- 16–18OctALZA Festival horror microdrama programme5days away
- 16–18OctALZA Festival in New York City5days away
- 16–17OctAustin AI Film Festival 20265days away
- 17OctFuture of Filmmaking Summit in Los Angeles6days away
- 19OctElevenLabs discount ends8days away
- 20–21OctRoboBusiness 20269days away
- 20OctAI-use disclosure required for applicants9days away
- 21OctLD Micro Main Event presentation10days away
- 21OctTesla reports third-quarter earnings10days away
- 23OctSanxingdui: Past and Future opens in cinemas12days away
- 24OctAI Film Summit Los Angeles 202613days away
- 24OctTyrannus Angel Awards 2026 winners announced13days away
- 26–29OctIBM TechXchange2weeks away
- 26OctFESTIAV AI film festival opens in Valencia2weeks away
- 27OctMistral plans to release Le Chonk weights2weeks away
- 30OctMicro-Series Challenge entry deadline3weeks away
- 30OctTesla's deadline to answer regulator questions3weeks away
- 31OctRegistration closes for AI Film Making course3weeks away
- 1NovSpeed Insights stops recording First Input Delay3weeks away
- 5–7NovThird LA Vertical Drama Market4weeks away
- 7NovTerry McMillan Presents: Paradise with You premieres4weeks away
- 9–11NovThe Hinton Lectures on AI safety4weeks away
- 21NovInaugural edition in Hollywood6weeks away
- 30–3Nov–DecContent London 20267weeks away
- 30NovRose d’Or Awards7weeks away
- 4DecAI-made film opens in limited release8weeks away
- 4DecGods Don’t Give Gifts premiere8weeks away
- 4DecSingapore AI Safety Fellowship ends8weeks away
- 12DecKampala screening9weeks away
- 14DecAI immuno-oncology primer at ESMO congress9weeks away
Stories are summarized in plain language; follow the link before quoting a figure. Story images are shown from the publishers' own sites.









































































