Add source Login
Why MongoDB thinks better retrieval — not bigger models — is the key to trustworthy enterprise AI

Why MongoDB thinks better retrieval — not bigger models — is the key to trustworthy enterprise AI

Agentic systems and enterprise search depend on strong data retrieval that works efficiently and accurately. Database provider MongoDB thinks its newest embeddings models help solve falling retrieval quality as more AI systems go into production. As agentic and RAG systems move into production, retrieval quality is emerging as a quiet failure point — one that can undermine accuracy, cost, and user trust even when models themselves perform well. The company launched four new versions of its embeddings and reranking models . Voyage 4 will be available in four modes: voyage-4 embedding, voyage-4-large, voyage-4-lite, and voyage-4-nano. MongoDB said the voyage-4 embedding serves as its general-purpose model; MongoDB considers Voyage-4-large its flagship model. Voyage-4-lite focuses on tasks requiring little latency and lower costs, and voyage-4-nano is intended for more local development and testing environments or for on-device data retrieval. Voyage-4-nano is also MongoDB’s first open-weight model. All models are available via an API and on MongoDB’s Atlas platform. The company said the models outperform similar models from Google and Cohere on the RTEB benchmark. Hugging Face’s RTEB benchmark puts Voyage 4 as the top embedding model. “Embedding models are one of those invisible choices that can really make or break AI experiences,” Frank Liu, product manager at MongoDB, said in a briefing. “You get them wrong, your search results will feel pretty random and shallow, but if you get them right, your application suddenly feels like it understands your users and your data.” He added that the goal of the Voyage 4 models is to improve the retrieval of real-world data, which often collapses once agentic and RAG pipelines go into production. MongoDB also released a new multimodal embedding model, voyage-multimodal-3.5, that can handle documents that include text, images, and video. This model vectorizes the data and extracts semantic meaning from the tables, graphics, figures, and slides typically found in enterprise documents. Enterprise’s embeddings problems For enterprises, an agentic system is only as good as its ability to reliably retrieve the right information at the right time. This requirement becomes harder as workloads scale and context windows fragment. Several model providers target that layer of agentic AI. Google’s Gemini Embedding model topped the embedding leaderboards, and Cohere launched its Embed 4 multimodal model , which processes documents more than 200 pages long. Mistral said its coding-embedding model, Codestral Embedding , outperforms Cohere, Google, and even MongoDB’s Voyage Code 3. MongoDB argues that benchmark performance alone doesn’t address the operational complexity enterprises face in production. MongoDB said many clients have found that their data stacks cannot handle context-aware, retrieval-intensive workloads in production. The company said it's seeing more fragmentation with enterprises having to stitch together different solutions to connect databases with a retrieval or reranking model. To help customers who don’t want fragmented solutions, the company is offering its models through a single data platform, Atlas. MongoDB’s bet is that retrieval can’t be treated as a loose collection of best-of-breed components anymore. For enterprise agents to work reliably at scale, embeddings, reranking, and the data layer need to operate as a tightly integrated system rather than a stitched-together stack.

Equal1, which was spun out from University College Dublin, raised $60M to deploy its new quantum server for data centers, bringing its total funding to $85M (Ciara O'Brien/The Irish Times)

Equal1, which was spun out from University College Dublin, raised $60M to deploy its new quantum server for data centers, bringing its total funding to $85M (Ciara O'Brien/The Irish Times)

Ciara O'Brien / The Irish Times : Equal1, which was spun out from University College Dublin, raised $60M to deploy its new quantum server for data centers, bringing its total funding to $85M —  UCD spinout has developed a quantum server for data centres  —  Equal1, the Irish quantum computing company that was spun …

ASUS has stopped producing the NVIDIA RTX 5070 Ti and 5060 Ti 16GB, saying they've reached 'end of life'

ASUS has stopped producing the NVIDIA RTX 5070 Ti and 5060 Ti 16GB, saying they've reached 'end of life'

YouTube channel Hardware Unboxed is reporting that NVIDIA has “effectively” discontinued the RTX 5070 Ti and 5060 Ti 16GB due to the ongoing memory crunch. In its most recent video , the channel states ASUS “explicitly” told it the RTX 5070 Ti is “currently facing a supply shortage.” As a result, the company has “placed the model into end of life status,” and no longer plans to produce it. Hardware Unboxed also spoke to retailers in Australia, who told the channel the 5070 Ti is “no longer available to purchase from partners and distributors,” adding they expect that to be the case throughout at least the first quarter of the year. The 5060 Ti 16GB “is almost done as well," with ASUS stating it no longer plans to produce that model going forward either. Both GPUs are 16GB models, making them more expensive to produce in the current economic climate. And while there might be some hope of the 5070 Ti and 5060 Ti 16GB returning later this year, the channel suggests both are unlikely to make a comeback. NVIDIA will reportedly focus on 8GB models like the RTX 5050 , 5060, and 5060 Ti 8GB , with the 12GB 5070 set to stick around for now. The 5080 and 5090 are seemingly safe as well, as more expensive, higher margin models, they offer more space for manufacturers to absorb component price increases. NVIDIA did not immediately respond to Engadget’s comment request. We’re also waiting to hear back from ASUS. We’ll update this article when the companies respond. The AI boom has created an insatiable demand for RAM and other computer components from data center infrastructure companies. In response, many memory manufacturers have shifted their production lines to focus on high bandwidth memory for those clients at the expense of their regular offerings, leading to dramatically increased prices among consumer RAM kits , GPUs and SSDs. In December, Micron Technology announced it would wind down its consumer-facing Crucial brand to focus exclusively on providing components to the AI industry. ASUS is the first of NVIDIA’s add-in board (AIB) partners to comment on the memory crunch. AIBs are the companies that produce the majority of GPUs you can buy from NVIDIA and AMD. Historically, NVIDIA has provided its board partners with both the GPU die and memory needed to make a graphics cards. However, a recent rumor suggested the company told partners they would need to start sourcing memory on their own. If this is in fact the demise of the 5070 Ti and 5060 Ti 16GB, it’s sad news for PC enthusiasts. Many modern AAA games demand more than 8GB of VRAM, making the 16GB GPUs from both NVIDIA and AMD the ones you want to buy if you’re building a new system or upgrading your current rig. Update 12:55PM ET: Added more context. This article originally appeared on Engadget at https://www.engadget.com/gaming/pc/asus-has-stopped-producing-the-nvidia-rtx-5070-ti-and-5060-ti-16gb-saying-theyve-reached-end-of-life-162012253.html?src=rss

Laptop makers embraced AI. Then Microsoft left them hanging

Laptop makers embraced AI. Then Microsoft left them hanging

Months ago, Microsoft announced that every Windows 11 PC would be an “AI PC” , even the non-Copilot+ ones. Then why was everyone pushing Copilot+ AI PCs at CES 2026? The industry finally caught up to Microsoft’s Copilot+ requirements—with a big NPU push from Intel in particular—but Microsoft didn’t explain why we should care. I saw a wave of Copilot+ PCs at CES 2026, but it felt like they were chasing an AI PC strategy that Microsoft has already abandoned. With Microsoft now downplaying NPUs and few applications taking advantage of them, the great NPU push doesn’t feel very important . That’s especially true since the Windows AI Foundry will use GPUs and CPUs for AI applications instead of NPUs, as the initial Copilot Runtime did. NPUs seem less necessary to the future of AI on Windows, even as they’re starting to pop up everywhere. Did Microsoft get distracted just as its PC hardware partners crossed the finish line? At CES 2026, NPUs finally feel fast enough When Microsoft unveiled Copilot+ PCs , the company required NPUs capable of at least 40 trillion operations per second (TOPS). This was a huge blow to Intel. Most Intel-powered machines have been shipping with NPUs capable of 13 TOPS at best, aside from Lunar Lake-powered machine with NPUs capable of 48 TOPS. 2024 was “ the year of the AI PC ,” but even throughout 2025 most laptops I reviewed couldn’t muster the specs needed for AI features on Windows 11. I spoke to PC manufacturer PR people who showed me the new versions of laptops I reviewed last year. “And it’s a Copilot+ PC,” they say proudly. It seems they’ve finally caught up to the requirements. The NPUs everyone’s talking about at CES Intel’s Core Ultra Series 3 (Panther Lake) hardware is the big advance at this year’s CES, given that Intel was so far behind on NPUs before. Core Ultra Series 3 has a 50 TOPS NPU and also promises big improvements to multithreaded performance, but we’ll have to run our own benchmarks to see just how big an upgrade it is in practice. While Intel’s Lunar Lake hardware was Copilot+ PC-capable, it was severely limited on multithreaded performance, which meant that an Intel laptop had no hope of running Copilot+ PC features unless you were willing to make big performance sacrifices and prioritize low power consumption and long battery life. Foundry / Mark Hachman AMD’s Ryzen AI 400 series hardware includes an NPU capable of 60 TOPS, and it’s coming to both laptops and desktop PCs. While AMD has been delivering capable NPUs for a while—unlike Intel-powered laptops—it’s an increase from the 50 TOPS NPUs in the Ryzen AI 300 series. However, with so few applications taking advantage of the NPU, that bump of 10 TOPS won’t be noticeable to the average laptop buyer, even if it looks like an upgrade on a spec sheet. Qualcomm is extremely proud of its TOPS speeds, highlighting that the Qualcomm Hexagon NPUs on Snapdragon X2 Elite and Snapdragon X2 Plus hardware deliver 80 TOPS of performance. Qualcomm’s Snapdragon X platform was the big launch partner for Copilot+ PCs, and Qualcomm is once again ahead. But as it was during the flashy Copilot+ PC launch, there still isn’t a great argument for NPUs just yet. All those new processor platforms are now delivering fully capable NPUs that will end up in laptops from all the big PC manufacturers. Going into 2026, Copilot+ PC-capable NPUs are finally becoming much more common. But will it matter? All Windows 11 PCs are now AI PCs Back in October, Microsoft revealed its plan to make every Windows 11 PC an AI PC . Here’s what Yusuf Mehdi told reporters at the time: “We did all of this years of work that let us get to the point of understanding what’s the right way to bring AI in. We’ve learned a lot from that—you know, what features resonate. And one of the big things that I think really came to us is, while Copilot+ PCs really are the tip of the spear and are gaining, you know, fast traction, the big thing was, let’s bring that AI capability to all Windows 11 PCs and make it really simple for anyone to try it. So, that has been the big thing.” As we turn the corner and head into 2026, it doesn’t sound like Microsoft is all that excited about NPUs anymore! And that’s without even mentioning the Windows AI Foundry . Developers can use it to write AI apps that perform inference on GPUs, CPUs, or NPUs. It replaces the Copilot Runtime, which required an NPU on Copilot+ PCs. Copilot+ PC features haven’t taken the world by storm I was in a Lyft earlier this week in Las Vegas. After asking me what I did for a living, the driver mentioned he was still using Windows 11. “There’s probably a newer version by now, right?” No, I explained: Windows 11 is still the latest version. You get some extra AI features if you have a newer PC—but only certain newer PCs. On those Copilot+ PCs, you don’t get extra Copilot features. Instead, you get minor features like Windows Studio Effects webcam effects, image generation in the Photos app, Windows Recall for searching your PC usage, and Click To Do for taking basic actions on text. Chris Hoffman / Foundry Based on Microsoft’s talk about delivering more AI features to all Windows 11 users, I don’t expect NPUs to become the crown jewel of the Windows AI experience in the future. If anything, I expect the opposite: I can picture a Windows 11 update that delivers Copilot+ PC features to a wider variety of machines, letting your PC’s GPU power features like image generation and text summarization. That’s what I hope to see. Microsoft should’ve never required NPUs for Copilot+ PC features. Even my $3,000 gaming PC still can’t run Copilot+ PC AI features , which is astonishing considering the fact that a speedy discrete GPU is still the best way to run more “serious” AI tools like LM Studio. Further reading: The 10 best laptops at CES 2026