Latest AI News
Can tech companies learn to love cheaper AI models?
The AI boom has been built on a basic assumption: Bigger models are more powerful, and the most powerful models win. Now, the industry is about to learn what happens if that assumption starts to break. Mounting costs have already pressured users to give smaller and cheaper models a second look. Thiscost-conscious model-shoppingis new and it’s unclear how it will affect the industry, but the impact is likely to be significant. One prediction, laid out best by Coinbase co-founder Brian Armstrong, is that it will result in the vast majority of tasks shifting to cheaper models. “[D]emand for intelligence is near infinite, but 80% of workloads will be running on 99% cheaper models within 12-18 months,” Armstrongwrote on X. “20% of workloads will still run on latest gen models where IQ maxing is important.” It’s hard to overstate what a significant shift it will be for the AI industry if Armstrong’s prediction comes true. Before now, most AI companies have competed on quality, which has meant defaulting to the most advanced available model. If those same jobs can be handled by cheaper models without affecting quality, it would mean a massive shift in the economics of AI. And critically, much of the savings would be coming out of the pockets of the big labs, dealing a financial blow to OpenAI and Anthropic just as they’re heading for their IPOs. It’s a potentially seismic change in the industry, resting on one basic question: Are companies ready to switch to smaller models? Initial tests suggest that, when the system is arranged right, cheaper models could sub in without any sacrifice in quality. In a recent test by the legal AI tool Harvey, the company was able to reduce inference costs by 3x without reducing quality. The test,performed in partnershipwith the inference platform Fireworks AI, combined Claude Opus and Fireworks’ GLM 5.1, and shifted to Opus for the most intensive tasks. The result was a significantly lower load in terms of server time and overall cost. “Quality comes first, and in legal it always will,” Harvey co-founder Gabe Pereyra told TechCrunch, referring to the AI legal services his startup provides. “However, the definition of quality is evolving from simply using the most powerful model for everything, to using the best model that gets the right answer most efficiently.” This trend is often framed in terms of major labs versus Chinese models or open-weight ones, but that misses the bigger point. The real divide isn’t between proprietary and open models; it’s between large models and small ones. You can save money by switching from GPT-5.5 to DeepSeek’s V4 Flash, but switching to GPT-5.4-mini works just as well. There’s an active price war going on between in-house inference from the big labs and independently served open-weight models. For the bigger question of small versus large, it doesn’t really matter which kind of small model wins out. All of this might seem obvious — of course you shouldn’t use more compute than necessary — but it runs counter to the scaling-first approach that has dominated the industry until now. Inspired bythe bitter lesson, labs have leaned hard into training the most compute-intensive models possible, pushing the frontier of what AI models can do. With prices heavily subsidized by investors, clients had no reason to choose anything but the most advanced option. With token prices rising and subsidies slowing down, users are facing cost pressure for the first time. We don’t know whether the new cost pressure will actually drive enterprise users to smaller models. They could just as easily economize by making fewer calls, using less context, or simply giving up on the least promising deployments. But if it turns out that most deployments can be run just as well on a smaller model, it could put a serious damper on the growing demand for inference — and raise new questions about how to justify the cost of training a frontier model.
View
Anthropic’s Fable 5 can make weirdly fun video games with the click of a button
Anthropic hasreleased Claude Fable 5, the first publicly available version of itsclosely watchedMythos model. What can Fable actually do? All kinds of things, it turns out. Ethan Mollick, a notable AI researcher and University of Pennsylvania scholar,has been playing aroundwith the model and seems to be having a lot of fun. In his testing, Fable consistently “outperformed basically every other public model I have used by a considerable margin,” Mollickwrote Tuesdayon his Substack. He added that it was “capable across many problems and produced some startling results — it would work up to a dozen hours executing on multi-page specifications.” Perhaps most strikingly, Mollick used Fable to create a variety of video games — all of which were generated via “one initial prompt” in Claude Code, the researcher says. Among these,Snakeis exactly what it sounds like. You’re a Pac-Man-like snake and you roam around eating apples. The snake never stops moving, and if you run off the screen, you die. It’s very 1980s arcade but, like many of those old games, it’s weirdly addicting. I played it longer than I’d like to admit before remembering I am a gainfully employed writer and not, in fact, a serpent who likes fruit. Then there wasStrata, where you’re roaming around in a seemingly endless network of subterranean tunnels and the goal is just to light as many lanterns as possible. The graphics look like a degraded version of Myst — they aren’t great — but the fact that the game exists at all, generated from a single prompt, is impressive. Mollick even managed to createDuino, a game based on the Duino Elegies, the celebrated cycle of poems by poet Rainer Maria Rilke. I like the animation here best — the player is a lone figure in a nocturnal landscape — although there isn’t much to the gameplay other than walking around while Rilke passages materialize on the screen. Aside from the variety of instant games Mollick produced, he also used Fable to create anisochronic map— a visualization showing how long it takes to travel between any two locations. The accuracy and detail is arresting. The implications are pretty clear. Software projects that once required entire teams — games, mapping tools, highly complex specifications — are now being spun up from a single prompt. It’s reason for vibe coders of the world to rejoice. As for founders and operators watching AI capability curves, it’s a useful data point about how quickly the floor is rising.
View
Hey Siri, here’s what I actually want from AI
Two years and a$250 million lawsuitlater, Apple’sAI Siri revampis on its way to your phones and laptops and even yourmixed reality headset, if you happen to be one of like three people who actually uses the Apple Vision Pro. Apple revealed a slew of new information at Monday’sWWDC keynoteabout these long-awaited, AI-powered updates that can take advantage of the fact that our hardware is supposedly “built for Apple Intelligence.” To be honest, it’s hard for AI to impress me enough that I’ll use it in my day-to-day life. I still don’t trust LLMs to provide consistently accurate information, I find it ethically untenable (and uncool) to use AI to help me write, and I don’t feel the insatiable urge to knowwhat I would look like as a Studio Ghibli character. But every once in a while, the promise of AI tempts me. That’s how I felt watching Apple’sSiri AI demos, which depict a world where your phone comes with an always-on, constantly-working assistant who knows everything about you and can help you keep track of all of the conversations happening on like 12 different apps on your phone at any given moment. To paraphrase Katy Perry, it feels so wrong (what are the privacy implications?), but it also feels so right (I am so overwhelmed by my phone and am begging for help parsing it all). I want Siri to be my own personal Emily from “The Devil Wears Prada” — a “second brain” that anticipates my needs before I even know what they are. I want Siri to read my texts and automatically make an event when a friend and I decide we’re going to meet up for dinner on Thursday. I want Siri to remind me when I’m walking past CVS that I have a prescription ready for pickup. If I forget to reply to an important work email, I want Siri to remind me that I didn’t write back yet. Siri AI won’t be able to do all of that out of the box, but it’s moving in the right direction. In one example at WWDC, Justin Titi, an Apple senior director working on AI engineering, asks the smart assistant to remind him of the dessert that his daughter mentioned recently. Siri searches across Titi’s phone to find a text from about a month ago, when his daughter mentioned that she wanted to make coconut cookies. It’s simple, but asking Siri to find that message saves time, rather than scrolling up through an entire month of conversation looking for that one specific text. The new-and-improved Siri is designed to use “personal context,” which refers to any information you put into Apple-native apps, like iMessage, Notes, Calendar, Mail, Photos, and more. Siri will also be aware of what’s on your screen, so for example, if you scroll past a picture of a nice park on Instagram, you can ask it to find out where that park is. (We still don’t know if Siri will be able to integrate into non-native Apple apps; it seems like it might be up to the developers to make that happen.) There already are apps likePoppyandPokethat try to create this kind of mobile, agentic AI. But the paradox of these AI personal assistant tools is that you have to give up a lot of personal data and privacy to make them work correctly, which may just cause you more trouble (remember that time when aMeta researcherranOpenClawand accidentally deleted her entire inbox?). I can’t say that I love giving any tech giant my personal data, but Apple at least seems to care more about security than the other FAANG (MANGOS?) companies. On-device AI will always be more secure and less energy intensive than cloud computing, since the data is processed directly on your phone. (This is how current Apple Intelligence features like email summaries and AI emojis are generated.) But for the more complex tasks that Siri will confront, Apple pioneeredprivate cloud compute(PCC), a way for devices to parse complex data over the cloud without even exposing your data to Apple itself. (If it’s possible to hack PCC, it hasn’t happened yet, even though Apple offers a$1 million bug bounty.) In a recent conversation with the writerCalvin Kasulke— who is so internet-brained that hewrote a novel that takes place exclusively on Slack— I confessed what feels like a taboo desire to outsource all of my “life admin” to an AI. “When you talk about the nonsense of the tech detritus in your life… I think the question is, ‘Is all that you have necessary?’ If it is necessary, isn’t it worth cultivating the skill and spending the time to do it?” Calvin told me. “I don’t think that those are skills that one should allow to atrophy.” He makes a good point: Maybe instead of asking Siri to remind me about the TV show that my friend told me I should watch, I could pay more attention when I’m talking to my friends. I don’t want to get into the habit of forgetting more consequential details from my conversations. “I’m sorry, but all of the commercials that are like, ‘What if I had the computer buy my kid a birthday gift?’ I’m like, ‘What if you learned what your kid likes?’ … Like, I don’t know man, it sounds like [they] don’t want to do the fundamental act of being a person,” he said. Maybe when I say I want Siri to be like Emily from “The Devil Wears Prada,” I should remember that Emily’s character is on the verge of a crash-out. I know I can’t psychologically impact Siri like Miranda Priestly damaged Emily, but will I become the kind of person who can’t function without the friendly robot voice in my phone? Do I want to be that person? At least if I decide to opt out from all of this, Apple will make that possible. Unlike Google’s controversialSearch overhaul, the new AI Siri can be toggled on and off, so you don’t have to use it. Until then, I’ll have to decide if it’s worth it to taste the forbidden fruit of Siri AI.
View
Anthropic Releases Claude Fable 5, Its First Mythos-Class Model for General Users
Anthropic is also introducing Claude Mythos 5, a version of the model with some safeguards removed for select cybersecurity and life sciences researchers.
View
It’s not FAANG anymore. It’s MANGOS.
WithSpaceX about to break recordswith an IPO on Friday,Anthropic about to break recordswith its pending IPO, andOpenAI racing to match or best its archrivalswith its own potentially record-breaking IPO, the tech industry will soon have a new set of public company overlords. Should all these IPOs take place as planned, these companies will be replacing the vicious-soundingFAANGcabal — Facebook (now Meta), Amazon, Apple, Netflix, Google (now Alphabet) — with the delightfully sweet-sounding (though truly sour and atrocious if consumed unripe) coterie MANGOS: Meta, Anthropic, Nvidia, Google, OpenAI, SpaceX. As these companies go, so shall the whole tech industry, or so it’s looking like from the summer of 2026. The new acronym was proposed by developer@krishdotdevand@lilscooton X and is now going viral. Of course, FAANG is not exactly dead — Amazon and Netflix remain powerful — but streaming services and Amazon’s e-commerce business, if not its cloud, are perhaps less groundbreaking these days than the AI and agentic companies the tech industry is about to crown. To that we say: Farewell to FAANG! Long live the MANGOS! (At least if they prove to be a nourishing foundation of a healthy economy powered by an upcoming autonomous AI age, and don’t usher in an unpalatable future where we all wind up jobless and broke.) i prefer MANGOSpic.twitter.com/fNnomeqgL5
View
Anthropic’s Claude Fable 5 is a version of Mythos the public can access today
Anthropic is bringing its most powerful AI model to the general public for the first time, but it’s doing it with guardrails. On Tuesday, the AI firm launched Claude Fable 5, the first publicly available version of its Mythos model. Anthropic says Fable 5 excels at software engineering, knowledge work, and vision, but it comes with hard safety limits. In high-risk areas like cybersecurity, biology, chemistry, anddistillation, the model blocks responses and falls back to Claude Opus 4.8. Launched as a preview in April, Mythos was initially limited to a handful of partners due to cybersecurity concerns. Last week, Anthropicexpanded access to hundreds of organizationsacross 15 countries, again focusing on organizations that manage critical infrastructure. Now, a version of that technology is available to anyone through Anthropic’s Claude API and consumption-based Enterprise plans. Access on subscriptions will roll out in stages: through June 22, Fable 5 is be included in Pro, Max, Team, and seat-based Enterprise plans at no extra cost. On June 23, Anthropic will pull Fable 5 from those plans, requiring usage credits going forward, with plans to restore it as a standard subscription feature as soon as possible. Anthropic is also deploying a new version of Mythos, called Mythos 5, to organizations that have already been approved to access the advanced model. Fable’s launch comes as Anthropic prepares to enter the public markets, alongsideOpenAIand Elon Musk’sSpaceX. It also follows theAI firm’s pleaurging major global AI labs to establish a coordinated brake pedal on frontier AI development. Anthropic warned that systems are advancing so rapidly that they may soon achieve recursive self-improvement (RSI), autonomously improving themselves without human intervention. Wary of what a Mythos-class model could do in the wrong hands, Anthropic says it stress-tested its classifiers with jailbreak attempts before releasing Fable 5. “Internally, we ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing. We then worked with external red-teaming orgs which also failed to find universal jailbreaks.” That said, there could still be novel attacks remain possible. As a result, with the launch of Fable 5 and Mythos 5, Anthropic said it will require a 30-day retention on all traffic, even if enterprises previously had zero-retention agreements. Anthropic said it won’t use the data for training, only to “defend against complex and novel attacks, including new jailbreaks,” and “identify and reduce false positives.” The policy could set an industry precedent in which access to increasingly powerful models comes with mandatory data retention policies framed as a safety measure. For those that continue to use the model, not every question will get a Fable 5 answer. Anthropic says the cases in which Fable has to defer to Opus 4.8 are rare, with early data showing at least 95% of Fable sessions running entirely on the model’s own responses. In third-party testing, analytics company Hex said in a statement that Fable was the first to get a 90% on its core analytics benchmark of complex, long-running analytical tasks. “On the hardest questions, it shows strong judgement and attention to nuance,” Hex said. Vibe-coding platform Base44 noted in a statement that Fable is better at “one-shotting full apps” and has excellent tool-calling. AI-powered workspace and agent platform Genspark said Fable beat every other model in its evaluations, and performed significantly better on tasks like UI design and game coding. Pricing for both Fable 5 and Mythos 5 is $10 per million input tokens and $50 per million output tokens, double the price of Opus 4.8. That price alone might serve as a deterrent for widespread use. Many enterprises are growing critical of AI costs afterseeing the bills come inor blowing through their yearly AI budgets early. Advanced models like Opus 4.8 can exacerbate those issues, with advanced reasoning skills that can split a single request into multiple tasks. Anthropic said it expects demand for Fable 5 to be very high and difficult to predict. And indeed some, like shopping rewards platform Rakuten, might think the upside is worth the price point. “At the highest effort, Fable reflects on and validates its own work,” Rakuten said in a statement. “For us, that’s what makes highly autonomous operations possible — the extra thinking pays for itself.”
View
Zoho’s Sridhar Vembu Calls Salesforce a ‘Garbage Bin’
Zoho’s Chief Scientist questioned Salesforce’s long-term pricing model and urged buyers to seek multi-year pricing commitments before switching.
View
How an e-scooter founder raised $5 million to build space data centers
Here’s one metric for tracking SpaceX’s IPO later this week: The company has changed the venture industry’s perspective on long-term, capital-intensive space so much that a talented founder with no space experience can fund a space data center company. Orbital, a new firm that emerged in May from a16z’s startup accelerator program Speedrun with a $5 million seed round, is the latest company promising to do inference in space — just as soon as Starship is flying regularly. Other investors include Basis Set, Human Element, Wayfinder, Antler, Anti Fund, Ascent, Rubik, Zero Knowledge Ventures, LYVC, Feld Ventures, New Legacy, FNDR, UpHonest and Asterisk. Founder and CEO Euwyn Poon previously founded e-scooter company Spin in 2017 andsold it to Forda year later, joining the automotive giant. When he was ready to start a new company, a16z’s Speedrun was eager to get on board, according to partner Andrew Chen, who told TechCrunch that Poon worked through several ideas before landing on space data centers. You’re familiar with the pitch. There’s insatiable demand for AI compute, and deploying it is slow going on Earth. Why not head to space for limitless sunshine and limited environmental reviews? The main problem isthe brutal economicsof launching stuff into orbit, which currently leaves the business case unable to close. Orbital, like many of it competitors, is betting on SpaceX figuring out its Starship rocket and offering it to commercial customers. “We will get to full scale when Starship comes online,” Poon explained. The price of the Falcon 9, the current state of the art, “makes this not economically feasible.” For now, Poon and company — which includes about a dozen folks in Los Angeles, with experience at Amazon LEO, SpaceX, and Northrop Grumman — are working toward a demo flight that will see the company fly an Nvidia Blackwell chip on a partner’s satellite to test Orbital’s radiation shielding and thermal management tech. In 2028, the company hopes to launch its first data-processing spacecraft with Nvidia’s Space-1 Vera Rubin-class GPUs. At that point, the company wants to start doing piece-wise inference work, which would allow it to generate revenue with each satellite launched. That’s a similar path to rival data center start-upStarcloud, which already has a GPU in orbit and plans to launch several more to generate income until Starship enables them to deploy their full constellation. Orbital’s goal is to deploy 10,000 satellites that provide a distributed gigawatt of computing power, with each satellite providing 100 kw of power. For comparison, Elon Musk said SpaceX expects its AI satellites produce up to 150 kw, and Starcloud expects to field larger 200 kw-rated spacecraft to run chips. Some companies are too impatient to wait for Starship. Cowboy Space Company, another space data center startup backed by a16z, recently decided to startbuilding its own rockets. Jeff Bezos’ space company Blue Origin also announced plans to launch data centers into space using its New Glenn launch vehicle. Poon is confident that the breadth of AI demand will allow many companies to succeed. “There’s so many lanes for companies in our space to pursue,” he told TechCrunch, before rattling off an array of choices that included companies pursuing different AI workloads, designs, and concepts of what an space data center looks like. Chen said that Poon’s experience scaling up a company that deployed 250,000 scooters across 100 cities shows he can manage the tricky task of building an aerospace company. Over the long term, a project like this might take a decade and $5 billion or more, but Chen said venture firms are more comfortable with timelines like that. “This kind of thing would have sounded crazy 10 years ago when we were all building mobile apps,” he said. “Starting it in 2026 just lets you tap into all the energy and excitement that’s that’s happening in the capital markets.” Poon found his way into the space data center business by a circuitous route. After leaving Ford, he bought a Nvidia A100 on a lark, co-locating it in a Santa Clara data center and serving open-weight models. That first-hand experience convinced him the value in delivering compute in the era of AI. Now he’s just got to put a couple thousand GPUs in space.
View
Lovable says it has hit $500M in annualized revenue, with 1 million new projects a week
Europe’s fast-growing vibe coding startup, Lovable, tells TechCrunch it has surpassed $500 million in annualized revenue run rate. Lovable last discussed its revenue in February, when the companysaid it crossed $400 million. In August, 2024, Lovable said it could hit $1 billion in annualized revenue within 12 months. It may not be on track to double that figure by summer, but it is still reporting jaw-dropping growth; the company, founded in late 2023, hasn’t yet hit its three-year anniversary. The company also claims it has been used to build over 50 million projects and says usage has accelerated to one million new projects a week. According toa survey of those projectsthat run on the company’s blog, Lovable says its users are primarily non-technical, yet are increasingly building software they intend to monetize or use in their businesses. Its users are founders, designers, and salespeople building websites and e-commerce storefronts, as well as internal tools like CRMs, inventory systems, and HR platforms, the company says. That list tells a story. AI vibe coding platforms have been seen as a threat to legacy SaaS software. Why buy expensive annual contracts when you can just vibe code it yourself? Lovable’s survey appears to offer some data that this is indeed happening. Of course, Lovable — therefore most of the projects built on it — isn’t old enough to answer the harder question about vibe-coded software: will such an approach prove short-lived? That’s because it’s not the initial building part that’s the problem — it’s the maintaining part. Software operates almost like a living organism: even well-written, well-designed code that isn’t AI slop runs atop an ever-shifting stack of dependencies, third-party services, and infrastructure — all of which is constantly being updated, which means end-user software is always breaking. That’s why so many companies choose to buy instead of build. They want others to be responsible for keeping it running. We’ll have to see if Lovable and other vibe coders will transparently report abandoned projects as their platforms mature — aka the not-as-flattering stuff. If those abandonment rates are low, that will be thetrue indication that the so-called SaaSpocalypseis here and here to stay.
View
Sandstone raises $30M to bring AI to in-house legal teams
WithHarveyandLegoraburning through eight-figure funding rounds, legal tools have proven to be one of the fastest-growing and most hotly contested verticals among AI startups. But while those tools focus on private practice, some startups believe there’s still plenty of the legal market that isn’t being served. Sandstone, which announced $30 million in Series A funding on Tuesday, is focused on an overlooked slice of the legal space, focusing on the tangle of overlapping tasks and systems facing in-house legal teams. The Series A was led by Lightspeed Venture Partners, with participation from existing investors at Sequoia, Mantis VC, SV Angel, Operator Partners, Kearny Jackson, Daybreak Ventures, Litquidity Ventures, and others. The Series A comes just six months after a $10 million seed round in January, which was led by Sequoia. As the founders describe it, Sandstone’s initial user base will be the legal departments at small and mid-sized businesses. “They open up their laptop in the morning, they see all the work that’s come in through different intake channels, whether that’s Slack messages, emails, Jira,” co-founder and chief operating officer Jarryd Strydom told TechCrunch. “AI helps them route and triage that work appropriately, and then they can build custom workflows on top of our platform to actually execute work, whether that’s drafting, reviewing, or providing legal analysis.” The result has little in common with legal reasoning systems like Harvey and Legora. Instead, Sandstone focuses on relationship management and workflow automation, both tuned to the unique demands of in-house legal work. As Strydom sees it, the focus on in-house legal departments allows Sandstone to provide value where more generalized AI deployments often flounder. “One of the convictions of Lightspeed was that they really believe in highly specialized vertical AI,” Strydom says, “because it takes a granular understanding of workflows to really nail down how AI can help.” Sandstone will also face heated competition from frontier AI labs, which are increasingly turning their attention to the legal space. Anthropic has beensteadily expandingits Claude for Legal offering, adding new tools in May for case law searches and deposition prep.
View
Apple WWDC 2026: Every Major Announcement You Need to Know
From a smarter Siri to iOS 27 and macOS Golden Gate, here are all the major announcements Apple made at WWDC 2026.
View
Meesho’s AI Bet on Bharat Is Already Driving 75% of Its E-commerce Orders
While Google and OpenAI push into AI commerce, Meesho is betting that understanding Bharat’s shoppers will be its biggest advantage.
View
