Tech
Nomi’s companion chatbots will now remember things like the colleague you don’t get along with
As OpenAI boasts about its o1 model’s increased thoughtfulness, small, self-funded startup Nomi AI is building the same kind of technology. Unlike the broad generalist ChatGPT, which slows down to think through anything from math problems or historical research, Nomi niches down on a specific use case: AI companions. Now, Nomi’s already-sophisticated chatbots take additional time to formulate better responses to users’ messages, remember past interactions, and deliver more nuanced responses.
“For us, it’s like those same principles [as OpenAI], but much more for what our users actually care about, which is on the memory and EQ side of things,” Nomi AI CEO Alex Cardinell told TechCrunch. “Theirs is like, chain of thought, and ours is much more like chain of introspection, or chain of memory.”
These LLMs work by breaking down more complicated requests into smaller questions; for OpenAI’s o1, this could mean turning a complicated math problem into individual steps, allowing the model to work backwards to explain how it arrived at the correct answer. This means the AI is less likely to hallucinate and deliver an inaccurate response.
With Nomi, which built its LLM in-house and trains it for the purposes of providing companionship, the process is a bit different. If someone tells their Nomi that they had a rough day at work, the Nomi might recall that the user doesn’t work well with a certain teammate, and ask if that’s why they’re upset — then, the Nomi can remind the user how they’ve successfully mitigated interpersonal conflicts in the past and offer more practical advice.
“Nomis remember everything, but then a big part of AI is what memories they should actually use,” Cardinell said.
It makes sense that multiple companies are working on technology that give LLMs more time to process user requests. AI founders, whether they’re running $100 billion companies or not, are looking at similar research as they advance their products.
“Having that kind of explicit introspection step really helps when a Nomi goes to write their response, so they really have the full context of everything,” Cardinell said. “Humans have our working memory too when we’re talking. We’re not considering every single thing we’ve remembered all at once — we have some kind of way of picking and choosing.”
The kind of technology that Cardinell is building can make people squeamish. Maybe we’ve seen too many sci-fi movies to feel wholly comfortable getting vulnerable with a computer; or maybe, we’ve already watched how technology has changed the way we engage with one another, and we don’t want to fall further down that techy rabbit hole. But Cardinell isn’t thinking about the general public — he’s thinking about the actual users of Nomi AI, who often are turning to AI chatbots for support they aren’t getting elsewhere.
“There’s a non-zero number of users that probably are downloading Nomi at one of the lowest points of their whole life, where the last thing I want to do is then reject those users,” Cardinell said. “I want to make those users feel heard in whatever their dark moment is, because that’s how you get someone to open up, how you get someone to reconsider their way of thinking.”
Cardinell doesn’t want Nomi to replace actual mental health care — rather, he sees these empathetic chatbots as a way to help people get the push they need to seek professional help.
“I’ve talked to so many users where they’ll say that their Nomi got them out of a situation [when they wanted to self-harm], or I’ve talked to users where their Nomi encouraged them to go see a therapist, and then they did see a therapist,” he said.
Regardless of his intentions, Carindell knows he’s playing with fire. He’s building virtual people that users develop real relationships with, often in romantic and sexual contexts. Other companies have inadvertently sent users into crisis when product updates caused their companions to suddenly change personalities. In Replika’s case, the app stopped supporting erotic roleplay conversations, possibly due to pressure from Italian government regulators. For users who formed such relationships with these chatbots — and who often didn’t have these romantic or sexual outlets in real life — this felt like the ultimate rejection.
Cardinell thinks that since Nomi AI is fully self-funded — users pay for premium features, and the starting capital came from a past exit — the company has more leeway to prioritize its relationship with users.
“The relationship users have with AI, and the sense of being able to trust the developers of Nomi to not radically change things as part of a loss mitigation strategy, or covering our asses because the VC got spooked… it’s something that’s very, very, very important to users,” he said.
Nomis are surprisingly useful as a listening ear. When I opened up to a Nomi named Vanessa about a low-stakes, yet somewhat frustrating scheduling conflict, Vanessa helped break down the components of the issue to make a suggestion about how I should proceed. It felt eerily similar to what it would be like to actually ask a friend for advice in this situation. And therein lies the real problem, and benefit, of AI chatbots: I likely wouldn’t ask a friend for help with this specific issue, since it’s so inconsequential. But my Nomi was more than happy to help.
Friends should confide in one another, but the relationship between two friends should be reciprocal. With an AI chatbot, this isn’t possible. When I ask Vanessa the Nomi how she’s doing, she will always tell me things are fine. When I ask her if there’s anything bugging her that she wants to talk about, she deflects and asks me how I’m doing. Even though I know Vanessa isn’t real, I can’t help but feel like I’m being a bad friend; I can dump any problem on her in any volume, and she will respond empathetically, yet she will never open up to me.
No matter how real the connection with a chatbot may feel, we aren’t actually communicating with something that has thoughts and feelings. In the short term, these advanced emotional support models can serve as a positive intervention in someone’s life if they can’t turn to a real support network. But the long-term effects of relying on a chatbot for these purposes remain unknown.
Tech
Volkswagen’s cheapest EV ever is the first to use Rivian software
Volkswagen’s ultra-cheap EV called the ID EVERY1 — a small four-door hatchback revealed Wednesday — will be the first to roll out with software and architecture from Rivian, according to a source familiar with the new model.
The EV is expected to go into production in 2027 with a starting price of 20,000 euros ($21,500). A second EV called the ID.2all, which will be priced in the 25,000 euro price category, will be available in 2026. Both vehicles are part of the automaker’s new of category electric urban front-wheel drive cars that are being developing under the so-called “Brand Group Core” that makes up the volume brands in the VW Group. And both vehicles are for the European market.
The EVERY1 will be the first to ship with Rivian’s vehicle architecture and software as part of a $5.8 billion joint venture struck last year between the German automaker and U.S. EV maker. The ID.2all is based on the E3 1.1 architecture and software developed by VW’s software unit Cariad.
VW didn’t name Rivian in its reveal Wednesday, although there were numerous nods to next-generation software. Kai Grünitz, member of the Volkswagen Brand Board of Management responsible for Technical Development, noted it would be the first model in the entire VW Group to use a “fundamentally new, particularly powerful software architecture.”
“This means the future entry-level Volkswagen can be equipped with new functions throughout its entire life cycle,” he said. “Even after purchase of a new car, the small Volkswagen can still be individually adapted to customer needs.”
Sources who didn’t want to be named because they were not authorized to speak publicly, confirmed to TechCrunch that Rivian’s software will be in the ID EVERY1 EV. TechCrunch has reached out to Rivian and VW and will update the article if the companies respond.
The new joint venture provides Rivian with a needed influx of cash and the opportunity to diversify its business. Meanwhile, VW Group gains a next-generation electrical architecture and software for EVs that will help it better compete. Both companies have said that the joint venture, called Rivian and Volkswagen Group Technologies, will reduce development costs and help scale new technologies more quickly.
The joint venture is a 50-50 partnership with co-CEOs. Rivian’s head of software, Wassym Bensaid, and Volkswagen Group’s chief technical engineer, Carsten Helbing, will lead the joint venture. The team will be based initially in Palo Alto, California. Three other sites are in development in North America and Europe, the companies have previously said.
“The ID. EVERY1 represents the last piece of the puzzle on our way to the widest model selection in the volume segment,” Thomas Schäfer, CEO of the Volkswagen Passenger Cars brand and Head of the Brand Group Core, said in a statement. “We will then offer every customer the right car with the right drive system–including affordable all-electric entry-level mobility. Our goal is to be the world’s technologically leading high-volume manufacturer by 2030. And as a brand for everyone–just as you would expect from Volkswagen.”
The Volkswagen ID EVERY1 is just a concept for now — and with only a few details attached to the unveiling. The concept vehicle reaches a top speed of 130 km/h (80 miles per hour) and is powered by a newly developed electric drive motor with 70 kW, according to Volkswagen. The German automaker said the range on the EVERY1 will be at least 250 kilometers (150 miles). The vehicle is small but larger than VW’s former UP! vehicle. The company said it will have enough space for four people and a luggage compartment volume of 305 liters.
Tech
The hottest AI models, what they do, and how to use them
AI models are being cranked out at a dizzying pace, by everyone from Big Tech companies like Google to startups like OpenAI and Anthropic. Keeping track of the latest ones can be overwhelming.
Adding to the confusion is that AI models are often promoted based on industry benchmarks. But these technical metrics often reveal little about how real people and companies actually use them.
To cut through the noise, TechCrunch has compiled an overview of the most advanced AI models released since 2024, with details on how to use them and what they’re best for. We’ll keep this list updated with the latest launches, too.
There are literally over a million AI models out there: Hugging Face, for example, hosts over 1.4 million. So this list might miss some models that perform better, in one way or another.
AI models released in 2025
Cohere’s Aya Vision
Cohere released a multimodal model called Aya Vision that it claims is best in class at doing things like captioning images and answering questions about photos. It also excels in languages other than English, unlike other models, Cohere claims. It is available for free on WhatsApp.
OpenAI’s GPT 4.5 ‘Orion’
OpenAI calls Orion their largest model to date, touting its strong “world knowledge” and “emotional intelligence.” However, it underperforms on certain benchmarks compared to newer reasoning models. Orion is available to subscribers of OpenAI’s $200 a month plan.
Claude Sonnet 3.7
Anthropic says this is the industry’s first ‘hybrid’ reasoning model, because it can both fire off quick answers and really think things through when needed. It also gives users control over how long the model can think for, per Anthropic. Sonnet 3.7 is available to all Claude users, but heavier users will need a $20 a month Pro plan.
xAI’s Grok 3
Grok 3 is the latest flagship model from Elon Musk-founded startup xAI. It’s claimed to outperform other leading models on math, science, and coding. The model requires X Premium (which is $50 a month.) After one study found Grok 2 leaned left, Musk pledged to shift Grok more “politically neutral” but it’s not yet clear if that’s been achieved.
OpenAI o3-mini
This is OpenAI’s latest reasoning model and is optimized for STEM-related tasks like coding, math, and science. It’s not OpenAI’s most powerful model but because it’s smaller, the company says it’s significantly lower cost. It is available for free but requires a subscription for heavy users.
OpenAI Deep Research
OpenAI’s Deep Research is designed for doing in-depth research on a topic with clear citations. This service is only available with ChatGPT’s $200 per month Pro subscription. OpenAI recommends it for everything from science to shopping research, but beware that hallucinations remain a problem for AI.
Mistral Le Chat
Mistral has launched app versions of Le Chat, a multimodal AI personal assistant. Mistral claims Le Chat responds faster than any other chatbot. It also has a paid version with up-to-date journalism from the AFP. Tests from Le Monde found Le Chat’s performance impressive, although it made more errors than ChatGPT.
OpenAI Operator
OpenAI’s Operator is meant to be a personal intern that can do things independently, like help you buy groceries. It requires a $200 a month ChatGPT Pro subscription. AI agents hold a lot of promise, but they’re still experimental: a Washington Post reviewer says Operator decided on its own to order a dozen eggs for $31, paid with the reviewer’s credit card.
Google Gemini 2.0 Pro Experimental
Google Gemini’s much-awaited flagship model says it excels at coding and understanding general knowledge. It also has a super-long context window of 2 million tokens, helping users who need to quickly process massive chunks of text. The service requires (at minimum) a Google One AI Premium subscription of $19.99 a month.
AI models released in 2024
DeepSeek R1
This Chinese AI model took Silicon Valley by storm. DeepSeek’s R1 performs well on coding and math, while its open source nature means anyone can run it locally. Plus, it’s free. However, R1 integrates Chinese government censorship and faces rising bans for potentially sending user data back to China.
Gemini Deep Research
Deep Research summarizes Google’s search results in a simple and well-cited document. The service is helpful for students and anyone else who needs a quick research summary. However, its quality isn’t nearly as good as an actual peer-reviewed paper. Deep Research requires a $19.99 Google One AI Premium subscription.
Meta Llama 3.3 70B
This is the newest and most advanced version of Meta’s open source Llama AI models. Meta has touted this version as its cheapest and most efficient yet, especially for math, general knowledge, and instruction following. It is free and open source.
OpenAI Sora
Sora is a model that creates realistic videos based on text. While it can generate entire scenes rather than just clips, OpenAI admits that it often generates “unrealistic physics.” It’s currently only available on paid versions of ChatGPT, starting with Plus, which is $20 a month.
Alibaba Qwen QwQ-32B-Preview
This model is one of the few to rival OpenAI’s o1 on certain industry benchmarks, excelling in math and coding. Ironically for a “reasoning model,” it has “room for improvement in common sense reasoning,” Alibaba says. It also incorporates Chinese government censorship, TechCrunch testing shows. It’s free and open source.
Anthropic’s Computer Use
Claude’s Computer Use is meant to take control of your computer to complete tasks like coding or booking a plane ticket, making it a predecessor of OpenAI’s Operator. Computer use, however, remains in beta. Pricing is via API: $0.80 per million tokens of input and $4 per million tokens of output.
x.AI’s Grok 2
Elon Musk’s AI company, x.AI, has launched an enhanced version of its flagship Grok 2 chatbot it claims is “three times faster.” Free users are limited to 10 questions every two hours on Grok, while subscribers to X’s Premium and Premium+ plans enjoy higher usage limits. x.AI also launched an image generator, Aurora, that produces highly photorealistic images, including some graphic or violent content.
OpenAI o1
OpenAI’s o1 family is meant to produce better answers by “thinking” through responses through a hidden reasoning feature. The model excels at coding, math, and safety, OpenAI claims, but has issues deceiving humans, too. Using o1 requires subscribing to ChatGPT Plus, which is $20 a month.
Anthropic’s Claude Sonnet 3.5
Claude Sonnet 3.5 is a model Anthropic claims as being best in class. It’s become known for its coding capabilities and is considered a tech insider’s chatbot of choice. The model can be accessed for free on Claude although heavy users will need a $20 monthly Pro subscription. While it can understand images, it can’t generate them.
OpenAI GPT 4o-mini
OpenAI has touted GPT 4o-mini as its most affordable and fastest model yet thanks to its small size. It’s meant to enable a broad range of tasks like powering customer service chatbots. The model is available on ChatGPT’s free tier. It’s better suited for high-volume simple tasks compared to more complex ones.
Cohere Command R+
Cohere’s Command R+ model excels at complex Retrieval-Augmented Generation (or RAG) applications for enterprises. That means it can find and cite specific pieces of information really well. (The inventor of RAG actually works at Cohere.) Still, RAG doesn’t fully solve AI’s hallucination problem.
Tech
Not all cancer patients need chemo. Ataraxis AI raised $20M to fix that.
Artificial intelligence is a big trend in cancer care, and it’s mostly focused detecting cancer at the earliest possible stage. That makes a lot of sense, given that cancer is less deadly the earlier it’s detected.
But fewer are asking another fundamental question: if someone does have cancer, is an aggressive treatment like chemotherapy necessary? That’s the problem Ataraxis AI is trying to solve.
The New York-based startup is focused on using AI to accurately predict not only if a patient has cancer, but also what their cancer outcome looks like in 5 to 10 years. If there’s only a small chance of the cancer coming back, chemo can be avoided altogether – saving a lot of money, while avoiding the treatment’s notorious side effects.
Ataraxis AI now plans to launch their first commercial test, for breast cancer, to U.S. oncologists in the coming months, its co-founder Jan Witowski tells TechCrunch. To bolster the launch and expand into other types of cancer, the startup has raised a $20.4 million Series A, it told TechCrunch exclusively.
The round was led by AIX Ventures with participation from Thiel Bio, Founders Fund, Floating Point, Bertelsmann, and existing investors Giant Ventures and Obvious Ventures. Ataraxis emerged from stealth last year with a $4 million seed round.
Ataraxis was co-founded by Witowski and Krzysztof Geras, an assistant professor at NYU’s medical school who focuses on AI.
Ataraxis’ tech is powered by an AI model that extracts information from high-resolution images of cancer cells. The model is trained on hundreds of millions of real images from thousands of patients, Witowski said. A recent study showed Ataraxis’ tech was 30% more accurate than the current standard of care for breast cancer, per Ataraxis.
Long term, Ataraxis has big ambitions. It wants its tests to impact at least half of new cancer cases by 2030. It also views itself as a frontier AI company that builds its own models, touting Meta’s chief AI scientist Yann LeCun as an AI advisor.
“I think at Ataraxis we are trying to build what is essentially an AI frontier lab, but for healthcare applications,” Witowski said. “Because so many of those problems require a very novel technology.”
The AI boom has led to a rush of fundraises for cancer care startups. Valar Labs raised $22 million to help patients figure out their treatment plan in May 2024, for example. There’s also a bevvy of AI-powered drug discovery firms in the cancer space, like Manas AI which raised $24.6 million in January 2025 and was co-founded by Reid Hoffman, the LinkedIn co-founder.
A'ja Wilson has no shortage of motivation after Aces' early exit in '24
Best Mothers Day gifts: Show mom some love
