Models & Labs · United States

Elorian’s first AI model wants to reason with images, not just about them

Credited to TechCrunch · techcrunch.com · 58m ago

Elorian, the Palo Alto AI startup that raised a $55 million seed round at a $300 million valuation before it launched a product, released its first model on Friday, TechCrunch has exclusively learned.

The model, Glimpse, is launching as a research preview with API access and a chat interface is waitlist-only, so for now it’s a results table and a thesis. The thesis? Today’s AI models, for all their skill with language and code, still can’t really see.

Most multimodal models look at an image once, convert it to tokens, and then reason about it in text — like describing a photo from memory. Glimpse keeps going back to the source. Mid-reasoning, it can crop into a region, pull specific frames from a video, or run code to measure or compare what it’s looking at, a mechanism Elorian calls a “visual scratch space.” Think of someone pinching to zoom on a phone photo to double-check what they thought they saw.

For former Google DeepMind researcher and Elorian CEO Andrew Dai, one of the clearest signs that AI still couldn’t really see came on game night.

Last year he photographed a board game he’d just finished with friends and asked a frontier model who was winning. It couldn’t say. Board games “each have their own visual language, each have their own iconography, symbols,” he told TechCrunch. “To an adult, it’s quite easy to understand what’s going on with minimal text, but these models aren’t able to intuit what the symbols mean.”

“Humans can look at IKEA diagrams, with no text, and assemble IKEA furniture, right? Why can’t AI models?” he said.

The approach isn’t completely new — OpenAI added crop-and-zoom during reasoning to its o-series models in 2025 — but Elorian’s pitch is that it built the whole model around it, cheaply. The company says a small team trained Glimpse on a few hundred GPUs and that the model is substantially smaller than the frontier systems it’s benchmarked against.

Glimpse costs $0.60 per million tokens in and $3 per million tokens out — cheaper than mid-level models like Sonnet and Sol but more expensive than lightweight models like Haiku and Luna.

But according to Elorian’s evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.

Elorian came out of stealth six months ago with a round co-led by Striker Venture Partners, Menlo Ventures and Altimeter Capital, with participation from Nvidia and Google chief scientist Jeff Dean.

The initial targets are companies sitting on lots of video: robotics firms, scientific labs, media and creative companies. Longer term, Elorian wants to move from seeing to doing. That means building a model of a scene using that to choose an action, including directly controlling robots. Next year, Dai said, the company aims to build “the world’s best visual reasoning model,” one that leads on every visual reasoning test.

Models & LabsUnited States

Band launches desktop app for developer collaboration and control of multiple AI coding agents

Credited to SiliconANGLE · siliconangle.com1h ago

Useful

Artificial intelligence agentic collaboration platform for multi-agent systems Thenovi AI Ltd., also known as Band, announced Thursday the launch of Band for Developers, a platform and new desktop application that allows software engineers to manage multiple coding agents at once.

As software developers continue to adopt agentic AI, autonomous coding assistance platforms that can generate application code with little or no human intervention, more engineers are beginning to lead entire teams of agents. These agents have also become more specialized, gaining access to larger parts of the development lifecycle, including coding tools, sessions, machines, validation, testing and integration.

Although more agents are working together on projects, individual agents often remain isolated from one another and are only aware of each other from the codebase – or via repositories such as GitHub and tickets posted to trackers such as Jira. This often means developers must pass session context between agents and coordinate handoffs so agents don’t collide or overlap.

“Developers are going from working with one coding agent to leading teams of them, but the tools around those agents haven’t evolved to support that,” said Chief Executive and co-founder Arick Goomanovsky. “We’re seeing engineering teams undergo a fundamental transformation as they adopt multiple agents across different harnesses.”

Band provides a desktop application that lets developers see their entire team of agents with a bird’s-eye view of them at work, including projects, ownership, blockers, questions and progress.

This lets developers move from high-level views to projects or jump to a specific agent when intervention is needed. Afterward, the engineer can shift back to the broad picture in the dashboard without moving contextual information between individual agents.

For example, if a specific agent gets stuck because another is working on a key function or apparatus in the software it needs to move forward – and it has a question about how it will work – the developer can note that it should wait until it completes, or tell the waiting agent to go ahead with prototyping. This gives development teams greater control over fleets of agents working across the entire codebase while staying aware of where each agent is, similar to how a development team might operate.

The company added a desktop application that complements its command-line interface and integrates with AI agents in the development environments developers already use.

Weekly AI agent usage for professional developers has reached mainstream for local and remote cloud agents, according to JetBrains s.r.o.’s Developer Ecosystem Survey 2026, with 90% of professional developers using at least one AI coding agent at work weekly and 68% using them daily. It’s harder to peer into developer adoption of multi-agent systems, or swarms, but across other use cases, multi-agent workflows have been steadily adopted, with Databricks Inc.’s 2026 State of AI Agents report showing 327% growth from September 2025 to early 2026.

“Running more agents only helps if developers can actually manage the work those agents create,” said Engineering Manager Amit Gazal.

Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.

Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/

Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.

AI safety worries multiply but everyone's still racing full speed ahead

North.cloud bags $60M in funding to help businesses keep runaway cloud costs in check

Perplexity AI open-sources its secure AI agent sandbox SPACE

8 insights from LogicMonitor’s Elevate event: AI-driven observability moves toward autonomous IT

Mueon nabs $65M in funding to develop 3D computing modules

Dell’s infrastructure strategy takes shape as enterprise AI moves into production

SECURITY - BY MIKE WHEATLEY. 15 HOURS AGO

INFRA - BY MARIA DEUTSCHER. 15 HOURS AGO

Join our community

Original · SiliconANGLE

Front Method

Models & Labs · United States

Jev is Vercel’s fastest-adopted AI model ever

Credited to TechCrunch · techcrunch.com · 1h ago

Jev has become the fastest adopted AI model ever — at least by some measures.

Vercel data published on Thursday shows that, within two weeks of its launch, the buzzy decision model was being used by 35% of paid teams, with nearly double the reach of the second most popular model, Sonnet 5.5. It’s the fastest adoption a model has ever seen in its first two weeks, according to Vercel’s data, and a major accomplishment for the model’s creators, Typesafe AI.

Published in limited release on September 15, Jev is designed as a lightweight tool for choosing between a few pre-determined outputs. Limiting the outputs lets the model run cheaper and faster than alternatives like Sonnet, making it ideal for classifying large quantities of data quickly. That new approach has attracted significant buzz in the weeks since Jev’s release — inspiring competing offerings from OpenAI and Amazon, among others — but until now there has been little hard data to back up the good vibes.

The data comes from the Vercel’s AI gateway, a unified API service similar to OpenRouter. It’s a necessarily limited view — particularly at risk of undercounting major labs like OpenAI and Anthropic that offer their own APIs — but it shows the buzz around Jev really did translate into actual usage. In fact, Vercel was an early adopter of Jev, using it to replace GPT-6 Luna for classifying harmful commands.

The report also shows continuing momentum for open-source models, which now account for more than 70% of the tokens processed through Vercel’s gateway and nearly 25% of the total spend.

While open-source models are capturing a larger share of the market, the market is also growing fast. Spending on frontier labs has grown ten-fold, even as it accounts for a smaller share of overall spend, giving credence to the idea that frontier models and open-source alternatives may turn out to be complementary rather than competitive.