Techin' Geek

Many in the industry think the winners of the AI model market have already been decided: Big Tech will own it (Google, Meta, Microsoft, a bit of Amazon) along with their model makers of choice, largely OpenAI and Anthropic. 

But tiny 30-person startup Arcee AI disagrees. The company just released a truly and permanently open (Apache license) general-purpose, foundation model called Trinity, and Arcee claims that at 400B parameters, it is among the largest open-source foundation models ever trained and released by a U.S. company.

Arcee says Trinity compares to Meta’s Llama 4 Maverick 400B, and Z.ai GLM-4.5, a high-performing open-source model from China’s Tsinghua University, according to benchmark tests conducted using base models (very little post training).

Like other state-of-the-art (SOTA) models, Trinity is geared for coding and multi-step processes like agents. Still, despite its size, it’s not a true SOTA competitor yet because it currently supports only text.

More modes are in the works — a vision model is currently in development, and a speech-to-text version is on the roadmap, CTO Lucas Atkins told TechCrunch (pictured above, on the left). In comparison, Meta’s Llama 4 Maverick is already multi-modal, supporting text and images.

But before adding more AI modes to its roster, Arcee says, it wanted a base LLM that would impress its main target customers: developers and academics. The team particularly wants to woo U.S. companies of all sizes away from choosing open models from China. 

“Ultimately, the winners of this game, and the only way to really win over the usage, is to have the best open-weight model,” Atkins said. “To win the hearts and minds of developers, you have to give them the best.”

The benchmarks show that the Trinity base model, currently in preview while more post-training takes place, is largely holding its own and, in some cases, slightly besting Llama on tests of coding and math, common sense, knowledge and reasoning.

The progress Arcee has made so far to become a competitive AI Lab is impressive. The large Trinity model follows two previous small models released in in December: the 26B-parameter Trinity Mini, a fully post-trained reasoning model for tasks ranging from web apps to agents, and the 6B-parameter Trinity Nano, an experimental model designed to push the boundaries of models that are tiny yet chatty.  

The kicker is, Arcee trained them all in six months for $20 million total, using 2,048 Nvidia Blackwell B300 GPUs. This out of the roughly $50 million the company has raised so far, said founder and CEO Mark McQuade (pictured above, on the right). 

That kind of cash was “a lot for us,” said Atkins, who led the model building effort. Still, he acknowledged that it pales in comparison to how much bigger labs are spending right now.

The six-month timeline “was very calculated,” said Atkins, whose career before LLMs involved building voice agents for cars. “We are a younger startup that’s extremely hungry. We have a tremendous amount of talent and bright young researchers who, when given the opportunity to spend this amount of money and train a model of this size, we trusted that they’d rise to the occasion. And they certainly did, with many sleepless nights, many long hours.” 

McQuade, previously an early employee at open-source model marketplace HuggingFace, says Arcee didn’t start out wanting to become a new U.S. AI Lab: The company was originally doing model customization for large enterprise clients like SK Telecom. 

“We were only doing post-training. So we would take the great work of others: We would take a Llama model, we would take a Mistral model, we would take a Qwen model that was open source, and we would post-train it to make it better” for a company’s intended use, he said, including doing the reinforcement learning. 

But as their client list grew, Atkins said, the need for their own model was becoming a necessity, and McQuade was worried about relying on other companies. At the same time, many of the best open models were coming from China, which U.S. enterprises were leery of, or were barred from using. 

It was a nerve-wracking decision. “I think there’s less than 20 companies in the world that have ever pre-trained and released their own model” at the size and level that Arcee was gunning for, McQuade said. 

The company started small at first, trying its hand at a tiny, 4.5B model created in partnership with training company DatologyAI. The project’s success then encouraged bigger endeavors. 

But if the U.S. already has Llama, why does it need another open weight model? Atkins says by choosing the open source Apache license, the startup is committed to always keeping its models open. This comes after Meta CEO Mark Zuckerberg last year indicated his company might not always make all of its most advanced models open source. 

“Llama can be looked at as not truly open source as it uses a Meta-controlled license with commercial and usage caveats,” he says. This has caused some open source organizations to claim that Llama isn’t open source compliant at all.

“Arcee exists because the U.S. needs a permanently open, Apache-licensed, frontier-grade alternative that can actually compete at today’s frontier,” McQuade said.

All Trinity models, large and small, can be downloaded for free. The largest version will be released in three flavors. Trinity Large Preview is a lightly post-trained instruct model, meaning it’s been trained to follow human instructions, not just predict the next word, which gears it for general chat usage. Trinity Large Base is the base model without post-training.

Then we have TrueBase, a model with any instruct data or post training so enterprises or researchers that want to customize it won’t have to unroll any data, rules or assumptions.

Acree AI will eventually offer a hosted version of its general release model for, it says, competitive API pricing. That release is up to six weeks away as the startup continues to improve the model’s reasoning training.

API pricing for Trinity-Mini is $0.045 / $0.15, and there is a rate-limited free tier available, too. Meanwhile, the company still sells post-training and customization options. 

The plus-one passes at 50% off are almost sold out. Now the clock is really ticking.

In just 3 days, the best deal at the lowest ticket prices for TechCrunch Disrupt 2026 disappear. With demand already surging and early inventory moving fast, this is the final window to lock in record-low pricing and secure a plus-one for half the price while limited passes remain.

If Disrupt has been on your must-attend list, now is the time to save up to $680 on your pass and bring a plus-one at 50% off.

This pricing ends January 30, 11:59 p.m. PT, or the moment the remaining plus-one passes sell out. No extensions. No exceptions. Register now to save.

Why founders, VCs, and operators keep coming back to Disrupt

From October 13–15, Moscone West in San Francisco will become the global epicenter of tech. TechCrunch Disrupt is a curated, three-day experience built to maximize signal over noise, bringing together 10,000 founders, investors, operators, and tech leaders for 200+ expert-led sessions featuring 250+ influential voices.

Across the ecosystem, past attendees consistently point to the same value:

• Real access to founders, investors, and operators who are actively building.

• Conversations that turn into partnerships, funding, and hires.

• Practical insights you can apply immediately, not just inspiration.

• A clearer picture of where tech is headed before it hits the mainstream.

At Disrupt, you’ll explore what’s next as 300+ startups debut new breakthroughs, feel the high-stakes energy of the intense startup pitch-off in Startup Battlefield 200, and engage in curated, high-impact networking with the people shaping the future of tech.

Past Disrupt speakers

Keep an eye on the Disrupt 2026 event page for when the agenda goes live.

A more curated way to experience a tech event

Disrupt isn’t about wandering between sessions. It’s about intentional connections and curated experiences designed for how people actually grow in the tech industry.

Founders meet investors actively backing breakthrough ideas. VCs cut through the noise to discover startups aligned with their investment focus. Operators exchange real-world lessons on building, scaling, and shipping what’s next. Aspiring innovators get inspired with a front-row seat to tomorrow’s tech and invaluable insights from those shaping it.

If you’re hands-on in tech, Disrupt was built for you. Find your ticket match now to secure the lowest rate.

Unique passes designed for founders and investors

Founders and investors can unlock specialized passes designed to support your goals:

Founder Pass: Built to help you accelerate growth with the right insights, tools, and connections.

Investor Pass: Designed for discovering standout startups and expanding your portfolio through curated access.

Final call: plus-one passes are almost gone

The 50% off plus-one passes are nearly sold out, and this deal ends in just 3 days. Lock yours in before Friday, January 30 at 11:59 p.m. PT. Register now to save up to $680 on your TechCrunch Disrupt 2026 pass and bring a plus-one at 50% off while discounted passes remain.

The latest wave of AI excitement has brought us an unexpected mascot: a lobster. Clawdbot, a personal AI assistant, went viral within weeks of its launch and will keep its crustacean theme despite having had to change its name to Moltbot after a legal challenge from Anthropic. But before you jump on the bandwagon, here’s what you need to know.

According to its tagline, Moltbot (formerly Clawdbot) is the “AI that actually does things” — whether it’s managing your calendar, sending messages through your favorite apps, or checking you in for flights. This promise has drawn thousands of users willing to tackle the technical setup required, even though it started as a scrappy personal project built by one developer for his own use.

That man is Peter Steinberger, an Austrian developer and founder who is known online as @steipete and actively blogs about his work. After stepping away from his previous project, PSPDFkit, Steinberger felt empty and barely touched his computer for three years, he explained on his blog. But he eventually found his spark again — which led to Moltbot.

While Moltbot is now much more than a solo project, the publicly available version still derives from Clawd, “Peter’s crusted assistant,” now called Molty, a tool he built to help him “manage his digital life” and “explore what human-AI collaboration can be.” 

For Steinberger, this meant diving deeper into the momentum around AI that had reignited his builder spark. A self-confessed “Claudoholic”, he initially named his project after Anthropic’s AI flagship product, Claude. He revealed on X that Anthropic subsequently forced him to change the branding for copyright reasons. TechCrunch has reached out to Anthropic for comment. But the project’s “lobster soul” remains unchanged.

To its early adopters, Moltbot represents the vanguard of how helpful AI assistants could be. Those who were already excited at the prospect of using AI to quickly generate websites and apps are even more keen to have their personal AI assistant perform tasks for them. And just like Steinberger, they’re eager to tinker with it.

This explains how Moltbot amassed more than 44,200 stars on GitHub so quickly. So much viral attention has been paid Moltbot that it has even moved markets. Cloudflare’s stock surged 14% in premarket trading on Tuesday as social media buzz around the AI agent resparked investor enthusiasm for Cloudflare’s infrastructure, which developers use to run Moltbot locally on their devices.

Still, it’s a long way from breaking out of early adopter territory, and maybe that’s for the best. Installing Moltbot requires being tech savvy, and that also includes awareness of the inherent security risks that come with it.

On one hand, Moltbot is built with safety in mind: It is open source, meaning anyone can inspect its code for vulnerabilities, and it runs on your computer or server, not in the cloud. But on the other hand, its very premise is inherently risky. As entrepreneur and investor Rahul Sood pointed out on X, “‘actually doing things’ means ‘can execute arbitrary commands on your computer.’”

What keeps Sood up at night is “prompt injection through content” — where a malicious person could send you a WhatsApp message that could lead Moltbot to take unintended actions on your computer without your intervention or knowledge. 

That risk can be mitigated partly by careful setup. Since Moltbot supports various AI models, users may want to make setup choices based on their resistance to these kinds of attacks. But the only way to fully prevent it is to run Moltbot in a silo. 

This may be obvious to experienced developers tinkering with a weeks-old project, but some of them have become more vocal in warning users attracted by the hype: things could turn ugly fast if they approach it as carelessly as ChatGPT. 

Steinberger himself was served with a reminder that malicious actors exist when he “messed up” the renaming of his project. He complained on X that “crypto scammers” snatched his GitHub username and created fake cryptocurrency projects in his name, and he warned followers that “any project that lists [him] as coin owner is a SCAM.” He then posted that the GitHub issue had been fixed but cautioned that the legitimate X account is @moltbot, “not any of the 20 scam variations of it.”

This doesn’t necessarily mean you should stay away from Moltbot at this stage if you are curious to test it. But if you have never heard of a VPS — a virtual private server, which is essentially a remote computer you rent to run software — you may want to wait your turn. (That’s where you may want to run Moltbot for now. “Not the laptop with your SSH keys, API credentials, and password manager,” Sood cautioned.)

Right now, running Moltbot safely means running it on a separate computer with throwaway accounts, which defeats the purpose of having a useful AI assistant. And fixing that security-versus-utility trade-off may require solutions that are beyond Steinberger’s control.

Still, by building a tool to solve his own problem, Steinberger showed the developer community what AI agents could actually accomplish and how autonomous AI might finally become genuinely useful rather than just impressive.

In the wake of TikTok’s U.S. ownership change last week, some users are seeking out alternative platforms. One app gaining traction is UpScrolled, a social network that pledges to remain impartial to political agendas. The app currently ranks 12th overall in Apple’s App Store and second in the social networking category.

Upscrolled blends familiar features from Instagram and X, letting users share photos, videos, and text posts, discover new content, and send direct messages.

The app was founded last year by Issam Hijazi, a Palestinian-Jordanian-Australian technologist, with the aim of giving users a place to “freely express thoughts, share moments, and connect with others,” according to the app’s website. The team behind the app says they’re “building a platform that belongs to the people who use it — not to hidden algorithms or outside agendas.”

“UpScrolled is the foundation for a digital ecosystem that puts power back into the hands of the people — not the corporations,” Hijazi says in a statement on UpScrolled’s website. “It’s more than just an alternative to Meta, X, or TikTok — it’s a reimagining of what social media should be: a space where creators, communities, and businesses thrive independently, with real control, transparency, and accountability.”

The app, available on both iOS and Android, is grappling with a surge of new users, but says it’s scaling to keep up with the demand.

According to data from the market intelligence provider Appfigures, UpScrolled saw approximately 41,000 downloads between Thursday, the day the TikTok deal was finalized, and Saturday, accounting for nearly one-third of its lifetime installs. UpScrolled has seen an average of about 14,000 daily downloads since Thursday, representing a 2,850% increase in daily downloads.

In total, the app has been downloaded 140,000 times to date, with 75,000 of those being U.S.-based installs.

“Well, this is new… You showed up so fast our servers tapped out,” the company wrote in a post on Bluesky. “Frustrating? Yes. Emotional? Also yes. We’re a tiny team building what Big Tech stopped being. Right now we’re scaling on caffeine to keep up with what YOU started. Bear with us. We’re on it.”

The surge comes as TikTok announced last Thursday that it signed a deal with a group of non-Chinese investors to form a majority American-owned joint venture to keep the social app operating in the U.S. TikTok’s Chinese parent ByteDance owns less than 20% of the new entity, while the venture’s three managing investors, Oracle, private equity firm Silver Lake, and Abu Dhabi-based investment company MGX, each hold a 15% stake. Some users worry that TikTok’s new U.S. investors may have political allegiances to Trump.

Following the takeover, some users accused the app of potentially censoring certain political content. This criticism included a handful of high-profile users, including Senator Chris Murphy and singer Billie Eilish, who raised concerns that TikTok was suppressing or limiting posts criticizing ICE. Additionally, some said they were unable to search for information about ongoing protests in Minneapolis following the killing of Alex Pretti by border patrol agents.

However, TikTok attributed these issues to an ongoing data center outage, which has been impacting its app’s functionality.

Concerns also grew when TikTok released an updated privacy policy that allows the app to track users’ GPS coordinates, among other things. This led to some users encouraging people to delete the app in favor of alternatives, with UpScrolled being one of the popular choices, largely due to its promise to not shadowban anyone and give “every post a fair chance to be seen.”

UpScrolled isn’t the only app that’s seeing a surge in users following the takeover. Skylight, a TikTok alternative built on open source technology, says it has now topped over 380,000 sign-ups and is continuing to grow.

British startup Synthesia, whose AI platform helps companies create interactive training videos, has raised a $200 million Series E round of funding that brings its valuation to $4 billion — up from $2.1 billion just a year ago.

Unlike some other AI startups that are still a long way from turning a profit, Synthesia has found a lucrative business in transforming corporate training thanks to AI-generated avatars. With enterprise clients including Bosch, Merck, and SAP, the London-based company crossed $100 million in annual recurring revenue (ARR) in April 2025.

This milestone explains why Synthesia’s venture backers are literally doubling down. The Series E that nearly doubled its valuation was led by existing investor GV (Google Ventures), with participation from several other previous backers — including Series B lead Kleiner Perkins, Series C lead Accel, Series D lead New Enterprise Associates (NEA), NVIDIA’s venture capital arm NVentures, Air Street Capital, and PSP Growth. 

Aside from ongoing support, this round will bring both new and departing investors. On one hand, Matt Miller’s VC firm Evantic and the secretive VC firm Hedosophia are joining the cap table as new entrants. On the other hand, Synthesia will facilitate an employee secondary sale in partnership with Nasdaq, TechCrunch has learned.

To be clear, Synthesia isn’t going public just yet — Nasdaq isn’t acting as a public exchange in this operation, but as a private markets facilitator that will help early team members turn their shares into cash. These employee stock sales often happen outside of this framework, but usually at prices either below or above the company’s official valuation, and are sometimes frowned upon by other shareholders. With this process, all sales will be tied to the same $4 billion valuation as Synthesia’s Series E, while the company keeps an element of control.

“This secondary is first and foremost about our employees,” Synthesia CFO Daniel Kim told TechCrunch. “It gives employees a meaningful opportunity to access liquidity and share in the value they’ve helped create, while we continue to operate as a private company focused on long-term growth.”

For Synthesia, this long-term growth involves going beyond expressive videos and embracing the AI agents trend. According to a press release, the company is developing AI agents that will let its clients’ employees “interact with company knowledge in a more intuitive, human-like way by asking questions, exploring scenarios through role-play, and receiving tailored explanations.”

The company said early pilots have received positive feedback from customers, who reported higher engagement and faster knowledge transfer compared to traditional formats. This positive response explains why Synthesia now plans to make agents a “core strategic focus” to invest in, alongside further product improvements to its existing platform.

While it didn’t disclose revenue forecasts, the company hopes its platform will offer a welcome answer to the struggles of enterprises in keeping their workforce adequately trained despite rapid changes. “We see a rare convergence of two major shifts: a technology shift with AI agents becoming more capable, and a market shift where upskilling and internal knowledge sharing have become board-level priorities,” Synthesia’s co-founder and CEO Victor Riparbelli said in a statement.

Seeing boards care more about employees as a result of AI wasn’t on anyone’s bingo card, except perhaps Riparbelli. Together with his cofounder, Synthesia COO Steffen Tjerrild, Riparbelli took the initiative of conducting a secondary sale so that employees could share in the success of the unicorn company. Founded in 2017, Synthesia now has more than 500 team members, a 20,000-square-foot HQ in London, and additional offices in Amsterdam, Copenhagen, Munich, New York City, and Zurich.

While unusual for a British startup, this coordinated secondary sale isn’t a first and likely not a last, Synthesia’s head of corporate affairs and policy, Alexandru Voica told TechCrunch. “My guess is that as [U.K.-based] private companies stay private longer, this type of structured, cross-border employee liquidity may become increasingly common, so I wouldn’t be surprised to see others do it, either with Nasdaq or others,” he predicted.