Enterprises trying to feed PDFs, slides and scanned documents into AI pipelines keep running into the same wall: the tools either miss the structure — tables, charts, layout — or cost too much to run at scale.

Cohere released Parse 5 on Thursday, positioning it on price-to-performance, not raw accuracy — the right cost-capability mix for enterprise scale. Parse 5 is a 2.3-billion-parameter vision language model built to convert PDFs, slides and images into structured Markdown at enterprise scale. 

Cohere’s own published benchmark comparison puts Parse 5 behind three larger, general-purpose frontier models on accuracy. GPT-5.5, Opus 4.8 and Gemini 3.5 Flash all score higher than Parse on the three ParseBench dimensions Cohere reports. Cohere is not claiming the top score. It is claiming the best price for a score close to the top.The company priced the model at $1.50 per 1,000 pages through its API, with Model Vault, Cohere’s secure, single-tenant platform for managed inference, available for higher-volume deployment.

“Document parsing isn’t solved because the hard part isn’t reading text, it’s preserving structure and meaning,” Nils Reimers, VP of AI Search at Cohere, told VentureBeat. “Enterprise documents mix tables, diagrams, charts, and formatting that change the interpretation of the data. Most tools still drop structure or hallucinate content, and even frontier models break on layout‑heavy pages.”

Inside the single-pass architecture

Parse 5 takes a page as an image, runs it through a single vision-language model pass and returns structured Markdown, collapsing the OCR-plus-model pipeline most tools run as separate steps.

Architecture. It is a 2.3-billion-parameter vision language model built on Cohere Labs’ North-Micro-Vision-Instruct architecture, with an 8,192-token context window and roughly a 4.6-gigabyte footprint. It accepts a PDF, PowerPoint or JPEG page as a base64-encoded image and returns Markdown in reading order, with tables rendered as HTML, image descriptions and bounding box coordinates for tables and images.

Language coverage. Arabic, English, French, German, Italian, Japanese, Korean, Portuguese and Spanish get stable accuracy, with lower-accuracy zero-shot support elsewhere.

Output modes. The default output returns a Markdown string per page. A blocks mode returns typed elements, where each table carries its own HTML, bounding box and description, the format Cohere positions as what makes citation-level traceability possible for agents.

Availability. Parse 5 is generally available now through the Cohere API, Model Vault, Microsoft Foundry and AWS SageMaker.

The benchmark shows a trade-off, not a win

ParseBench is a benchmark that scores document-parsing tools against human-verified enterprise pages. Cohere reports Parse 5 scoring 79.2 across three dimensions: tables, content faithfulness and semantic formatting. That puts Parse 5 behind GPT-5.5 (84.4), Opus 4.8 (84.3) and Gemini 3.5 Flash (81.8), and ahead of LlamaParse’s Cost Effective tier (78.3), Mistral OCR 4 (74.5), Databricks AI Parse (72.4) and Azure Document Intelligence (69.3).

Cohere’s table notes two excluded dimensions, Layout and Chart, and attributes both to product scope rather than a performance gap. Parse 5 returns reading-order Markdown instead of per-element bounding boxes for text, and describes charts rather than extracting their underlying data, with chart-data extraction planned for a future version.

Reimers said that design choice reflects where agentic workflows actually break.

“For charts, for example, we provide a general description of the chart together with an indicator, how Agentic AI can visually inspect the chart,” Reimers explained. “Other solutions try to extract the data from the chart, but then miss out critical information (for example, the color or the pattern of a line) that leads to hallucinations in Chat and Agentic AI applications.”

Cost is where Cohere makes its real case. Reimers pointed to a workflow the company modeled for a large financial services firm.

“We ran the numbers for a large financial services workflow that processes 750 million documents a year and showed that choosing Parse 5 over a large general‑purpose model like GPT‑5.5 would reduce costs by more than 98 percent.”

That figure is Cohere’s own estimate for a single modeled workflow, not an audited deployment.

Where Parse 5 sits against the field

There is no shortage of options for enterprises looking at parsing solutions.

General-purpose frontier models, GPT-5.5, Opus 4.8 and Gemini 3.5 Flash, top the accuracy comparison but carry the cost and latency of a large model on every page. 

Then there are specialized parsers, including Mistral OCR 4, LlamaParse and open-weight options like Chandra OCR 2 and RedNote’s dots.mocr.

Hyperscaler document intelligence services, AWS Textract, Google Document AI, Azure Document Intelligence and Databricks AI Parse, compete more on ecosystem convenience than on raw parsing quality, and score lowest in Cohere’s own comparison.

Kevin Petrie, VP of Research at BARC US, said document parsing sits at the center of enterprise AI adoption right now.

“We’re completing a survey now that shows document analysis is by far the #1 use case for AI, with 62% adoption rates among organizations we polled,” Petrie told VentureBeat. “Documents and other unstructured objects, including images and so on, hold the proprietary context that organizations need to differentiate their agentic AI initiatives.”

Petrie added that only time will tell how Cohere’s cost-performance stacks up against frontier models, but strategically his view is that Cohere has the right focus. 

Stephanie Walter, Practice Leader for AI Stack at HyperFRAME Research, sees Cohere Parse 5 as sitting in a good spot between legacy OCR and using an expensive frontier model on every page. 

“Its potential advantage is delivering structure, spatial provenance and private deployment at a price suitable for high-volume ingestion,” Walter told VentureBeat. “It does not need to win every benchmark. It needs to make reliable enterprise-scale parsing economical.”

The real test is downstream, not on the benchmark

“Parsing is the first quality gate in the enterprise AI stack,” Walter said. “If tables, headings, images, or reading order are lost during ingestion, better embeddings and larger models cannot recover that missing structure.”

A benchmark score isn’t the only input that matters here. “Enterprises should test parsers against their own most difficult documents and measure downstream retrieval and task accuracy, not how clean the extracted text looks,” Walter said. “The right question is not ‘Did it read the PDF?’ but ‘Can the agent now use the information correctly?’”

Wisdom Audio is coming to CEDIA Expo 2026 with three new products aimed squarely at one of the fastest-growing and increasingly competitive segments of residential AV: high-performance loudspeakers and subwoofers designed to disappear into walls and ceilings or integrate cleanly with a room’s architecture. Custom installation audio has moved well beyond background music, with Monitor Audio, KEF, Sonus faber, Theory Audio Design, DALI, Lyngdorf Audio and other established high-end manufacturers offering increasingly sophisticated architectural speakers for stereo, multi-room audio and immersive home theater systems.

Against that increasingly crowded backdrop, Wisdom Audio will showcase three new CI products at CEDIA 2026: the Sage Series ICS3FR in-ceiling loudspeaker, the SUB2 passive subwoofer, and the Insight P3m series of customizable speakers and soundbars. The objective is not simply to hide the hardware; it is to deliver the kind of performance expected from a high-end music or home theater system while giving integrators considerably more freedom over placement, aesthetics and room design.

Wisdom’s ICS3FR, for example, uses the company’s planar magnetic technology in an architectural design, while the SUB2 was specifically developed to add low-frequency extension while maintaining a minimal visual footprint.

The ISC3FR

The ICS3FR is a “full-range” in-ceiling solution designed to provide an architecture-friendly, minimal-appearance installation while delivering high-resolution playback and matching 3-inch to 4-inch recessed round or square lighting fixtures.

The ICS3FR incorporates a Spiral Planar Magnetic Driver designed to provide ultra-wide dispersion, with a 120-degree circular coverage angle (-6 dB), and reproduce frequencies down to 40Hz using its 6.5-inch woofer. Its rated frequency response is 40Hz to 20kHz (+/-3 dB), with an F10 point of 30Hz (-10 dB), where meaningful bass output effectively ends. The ICS3FR requires no additional EQ or PEQ, only a high-pass crossover set between 35Hz and 40Hz.

The ICS3FR ships with everything required for 3-inch round or square and 4-inch round or square installations, as well as hard-surface and gypsum mounting. Custom-colored metal grilles and custom wood grilles are also available separately.

“The ICS3FR was created for projects where exceptional sound quality is essential, but the loudspeaker itself should virtually disappear into the architecture,” said Ron Rouse, Vice President of Sales and Marketing at Wisdom Audio. “At its heart is our Spiral Planar Magnetic Driver, which provides the ICS3FR with best-in-class off-axis dispersion, delivering wide, consistent coverage and exceptional clarity throughout the listening area. Combined with its full-range performance, minimal-appearance design, and flexible grille and mounting options, the ICS3FR gives integrators an ideal solution for projects where outstanding sound and refined aesthetics are equally important.”

SUB2 Minimal Appearance Subwoofer

The SUB2 is a passive subwoofer designed to complement the Sage Series ICS3FR minimal-appearance loudspeaker and is available as an optional addition. It provides additional low-frequency support, extending bass response down to 37Hz through a 3-inch opening.

Its compact, minimal-appearance design also makes the SUB2 well suited for in-wall installations and other space-constrained applications where conventional subwoofer placement may not be practical.

Wisdom Audio Insight P3m Series

Separate from the in-ceiling Sage Series ICS3FR and SUB2, Wisdom Audio is also showcasing its new Insight P3m Series of customizable soundbars and individual loudspeakers at CEDIA 2026.

The Insight P3m Series includes the passive P3m-V vertical mono, P3m-C horizontal mono, P3m-LR stereo, and P3m LCR configurations. Each model can be custom-built in lengths or heights of up to 100 inches (2,540mm), with custom color options also available.

The P3m models use a Spherical Planar Magnetic Driver for high frequencies, extending down to 900Hz, while 4.5-inch woofers handle the lower frequencies.

The P3m Series also incorporates distinctive crescent-shaped woofer apertures that function as acoustic lenses. These help smooth the front baffle to reduce tweeter diffraction while making the two woofers behave as though they are positioned closer together. The result is improved dispersion, reduced interference through the crossover region, and greater low-frequency extension from the woofers.

Behind each aperture sits Wisdom Audio’s patent-pending Dual Helmholtz Resonator, which uses two slightly different tuning points to control unwanted resonances more effectively despite normal manufacturing variations. 

These technologies help flatten the woofer response, reduce distortion, improve speaker-to-speaker consistency, and deliver more uniform tonal balance across the listening area.

“With the Insight P3m Series, we set out to create a loudspeaker that delivers the performance people expect from Wisdom Audio while giving integrators far greater design flexibility,” said Ron Rouse, Vice President of Sales and Marketing at Wisdom Audio. “Its customizable dimensions, color options, and multiple configurations allow the P3m to complement virtually any display or interior without compromising sound quality. We believe it is the ideal solution for clients who want exceptional performance in a beautifully integrated, highly versatile loudspeaker.”

Pro Tip: Detailed specifications have not been provided for the Insight P3m Series. The official product links (refer to our availability section) are expected to go live on August 26, 2026, which will include more detailed information and specifications.  Optional TV VESA mounts will be sold separately 

The Bottom Line 

Wisdom Audio’s new CEDIA 2026 lineup is aimed at integrators and homeowners who want high-performance audio without conventional speakers dominating the room. The ICS3FR and SUB2 stand out for delivering wide-dispersion, low-frequency-capable performance through extremely discreet ceiling openings, while the customizable P3m Series gives installers unusual flexibility in size, orientation, and finish. These are purpose-built CI products for premium stereo and home theater installations where sound quality, architectural integration, and aesthetics carry equal weight.

Price & Availability

The ICS3FR speaker and SUB2 are available through Authorized Dealers – consult dealer for pricing.  

The P3m-C, P3m-V,. P3m-LR, and P3m-LCR will be available through authorized dealers beginning August 26, 2026, starts in October 2026.

Related Reading:

Whoop has made screen-free wearables since its first proper release in 2015, but the shift in what Whoop considers itself to be about mirrors the change we see in the wider market. Today a Whoop band is made for just about anyone with the spare cash to pay for its $199 annual subscription. It’s a system to “help you improve how you sleep, train, and feel—starting day one,” to use Whoop’s own words. This couldn’t be further removed from the self-serious way Whoop described itself in 2015, as a “performance optimization system exclusively designed for elite athletes and teams.” What a change a decade makes.

Even if Whoop has always been grounded in the now-common concept of “training readiness,” things have altered drastically. So while it’s tempting to say Google’s Fitbit and company are playing catch-up with Whoop, that’s not quite it. It has homed in on the “low-distraction” lifestyle selling point just as the other bigger tech companies have.

On at least three fronts this can be attributed—or blamed—on AI, to some extent.

AI as the Good and Bad Guy

AI is a way for these wearable companies to upsell and make money, and on the other side, a major contributor to why a lot of us feel like we need low-distraction tech these days in the first place. While virtually every Whoop rival says it can be used without a subscription, some of these companies lean heavily on AI as an upsell.

Take Google and the Fitbit Air. Without a subscription, the app looks cute enough but is fairly spare. You get stats and charts, but if you want chatty explainers of what any of it means, you need Google Health Premium at $10 a month. Numbers and graph lines become paragraphs that reference specific quirks or events of the day because, to steal a line from humorist and podcaster John Hodgman, specificity is the soul of narrative.

And this is just the beginning. “The use of AI in monitoring and guiding health and wellness is still at the early stage of development but will likely become significantly more advanced as AI matures,” says Peter Richardson, vice president and cofounder of Counterpoint Research.

I find Google Fitbit’s chatbot ramblings quite annoying, with not-that-rare dips into the condescending, and I may even prefer the subscription-free experience. But my job has pushed me through the unenviable task of learning what VO2 Max actually means, what heart-rate variability trends indicate, and so on: not the norm.

AI could solve that long-standing issue with wearables—people glazing over data as soon as they see anything more than their daily step count. This AI use will be better represented on the phone than the watch screen, too, making the lack of screen a strength. It lets the AI shine.

Technostress

The most important influence of AI here is on the other side of the equation, though, as just the latest stage in the tech-ification of our daily lives. I’ve heard from multiple friends about how their recent work appraisals focused on higher-ups demanding greater AI use within their “personal workflow.” Perhaps you’ve had a similar experience.

Elizabeth Marsh, a doctoral researcher at the University of Nottingham who studies the dark side of the digital workplace, finds that digital tools consistently trigger a familiar cluster of harms: technostress, information overload, constant interruption, and blurred work–home boundaries.

Digital photos usually vanish into a folder the second you press the shutter. Maker Strange Inventions wanted a shot you could pull out of the camera, stand on a desk, and still overwrite later, so he built an instant camera whose film is a stack of tiny four-color e-paper cartridges.

You insert one of those small 3D-printed cartridges into the top of the compact body, press the red shutter button, and, assuming you’re within range, the Raspberry Pi 3 model A+ and the 5 megapixel camera module work their magic, reducing the scene to black, white, red, and yellow as the screen flickers to life before finally settling. Pinch the Game Boy-style release tab, slide out the cartridge, then insert a new one. E-paper has a nice side effect here in that it holds the image even when the battery is not in place, so each ‘film cartridge’ becomes a one-time, tangible photo that you may display or replace as needed.

Sale

Bambu Lab A1 mini 3D Printer + LED Lamp Kit, Ready-to-Use for Beginners

• A1 mini + LED Lamp Kit for Creative Light Projects: Bring your ideas to life with the included LED Lamp Kit. Simply print compatible lamp models and…
• The Perfect 3D Printer for Beginners: A1 mini 3D Printer is designed to make 3D printing easy from day one with automatic calibration, simple setup…
• Experience the Bambu Lab Ecosystem: Access MakerWorld’s huge library of ready-to-print models, manage prints through the Bambu Handy app, and enjoy…

All of the hardware is squeezed into a small 3D-printed enclosure, which includes a Raspberry Pi 3 A+, a Raspberry Pi Camera Module 3 Wide, the quirky 1.54-inch 200×200, four-color e-paper display, and a battery pack to make the whole thing portable. In the early attempts, a frail ribbon held the display in place, but he soon replaced it with pin headers, which serve the same purpose while also keeping the cartridge firmly locked in. Headers are low-cost and effective. He did have some difficulty finding something springy to do the same job at first, but the idea of employing pogo springs was too expensive…in the end. Headers are excellent value and also serve to keep the cartridge in place, which is exactly the effect they were hoping for if ‘film’ was to feel like film. The entire contraption comes together without the need for screws, however he later added a little grab tab to make pulling the old cartridge out less like wrestling a ‘Game Pak’.

When you press the shutter, you get to choose from only four colors, which is a good for the Python script that reduces the raw captures and dollops to just those four colors, then lets you play with contrast, brightness, saturation, sharpness, and dithering until the image looks like it was meant to be that way on that small screen. Holding down the shutter button allows you to switch between a dithered and flat look. The poor old plants ripped apart at the seams since there was no green on the panel, but the maker’s hand, a toy car, a tape measure, a keyboard, and a spool of printer filament fared much better. Some subjects require additional resolution; for example, a window view was murky at 200 pixels.

In the end, he kept shifting through the six cartridges until he found a set that he was satisfied with enough to write on the screen. For the time being, he is letting them alone, and they appear like miniature works of art on their exhibition stand. The Strange Inventions Lab shop sells project files for €24, which contain the FreeCAD source file, STL files, electrical wiring diagram, complete parts list, Raspberry Pi software, modified e-paper driver, an image-processing test suite, as well as a shopping list and some basic build notes to get you started. The tough part is that you still have to source the Raspberry Pi, camera module, and all of the other components yourself. He used largely off-the-shelf boards for this prototype, but if you really want it to be as compact as possible, he’s recommended a custom PCB as the next step. Six images now sit on his display stand, waiting to be dusted off and replaced another day…

Two sources also describe the inventory bot as too slow. One says Meta previously explored using drones as a faster alternative, but ultimately didn’t pursue the idea.

Inside the Testbed

The Altoona data center campus is often where Meta prototypes new technologies. “What they do here in Altoona, they take to their other sites and put that in place there,” says O’Connor.

Since last June, Meta has been trying out a pair of dual-armed Watney robots for some cabling work at one of its Altoona buildings, according to one of the people familiar with the company’s efforts. The robots are supervised by humans and can’t work as fast as people yet, but they show promise.

After Watney launched in 2023, it put out a demo of a laundry-folding robot that it claimed could also handle janitorial work inside businesses. The San Francisco startup quietly pivoted to data center robots early last year, according to archived copies of its website.

Meta is testing additional robots in New Albany, Ohio, at the company’s newest data center campus known as Prometheus. Four-wheel robots “with a scissor-lift-style riser and a six-axis arm on top” made by the automation giant ABB are being used there to reseat parts and, eventually, they may take on other tasks with diminishing human oversight, a person familiar with the product says.

But the idea that robots will prevail over humans isn’t guaranteed. “There’s been a lot of pilots and demos, but no provable working solution,” says Helen Oleynikova, CEO of Exclaim Robotics, which is developing prototypes to replace broken parts and reset cables in data centers.

At Meta’s data centers, some workers have complained about the amount of downtime the robots need to recharge their batteries, according to the people familiar with the matter. Meta has also acknowledged internally that robots aren’t equipped for certain tasks, such as the intense cabling work required to run an Nvidia GB300 supercomputer. “Things have been designed for human hands forever to make everything a five-minute repair,” says one former Meta employee. “Redesigning everything will take time.” But the robots are getting prepared.

This story originally appeared at wired.com.

A vacuum cleaner can’t get behind a radiator or blast cobwebs off a kitchen skylight, but Shark’s BlastBoss can, and it’s currently going for £89.10.

That £89.10 price tag marks a steep drop from the usual £129.99, a 31% discount worth £40.89 on this four-star cordless air blaster built to handle exactly the kind of mess a vacuum cleaner can’t manage alone.