London-based deep tech startup Stanhope AI has closed a €6.7 million ($8 million) Seed funding round to advance what it calls a new class of adaptive artificial intelligence designed to power autonomous systems in the physical world.

The round was led by Frontline Ventures, with participation from Paladin Capital Group, Auxxo Female Catalyst Fund, UCL Technology Fund, and MMC Ventures.

The company says its approach moves beyond the pattern-matching strengths of large language models, aiming instead for systems that can perceive, reason, and act with a degree of context awareness in uncertain environments.

Stanhope is developing what it terms a “Real World Model”, building on principles from neuroscience and computational theory to allow machines to learn and adapt on the fly.

“We’re moving from language-based AI to intelligence that possesses the ability to act to understand its world – a system with a fundamental agency,” says Professor Rosalyn Moran, CEO and co-founder of Stanhope AI.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Her team includes theoretical neurobiologist Professor Karl Friston, whose work on the Free Energy Principle informs the startup’s methodology.

Rather than relying on cloud-centric deep learning, Stanhope’s models are designed to run efficiently on edge devices with limited data and power. That fits a broader industry shift toward on-device AI, where systems must operate reliably in dynamic settings such as autonomous vehicles, robots, and defence hardware.

The firm says its technology is already being tested on drones and other autonomous platforms with international partners.

Stanhope’s funding comes amid sustained investor interest in AI and autonomy startups across Europe. In recent months, companies from robotic manufacturing to defence software have attracted capital, underscoring demand for systems that go beyond conventional machine learning.

Frontline Ventures partner Zoe Chambers said Stanhope’s progress from academic research to production-ready systems was a rare combination in the industry, and that the technology had clear potential in domains where machines must react and adapt in real time.

Christopher Steed of Paladin Capital Group highlighted the relevance of adaptive AI for critical and security-sensitive applications.

Founded in 2023 as a spin-out from University College London and King’s College London, Stanhope AI aims to carve a niche at the intersection of robotics, industrial automation, and defense.

The new capital will help push its technology further into real-world deployments, where adaptability and resilience are often the key barriers to broader adoption.

What to do with old tech can be a bit of a pain, especially if said gadgets are from premium brands like Apple. While it can be handy to have a spare secondary iPhone or iPad to hand in case a newer model goes wrong, sometimes such devices can sit in a drawer or cupboard and just gather dust.

Trading in Apple items to get a newer version is one way around that, but the trade-in values aren’t always great. But as we approach the Presidents’ Day sales, Best Buy is running an ‘Apple Trade-up Event’, offering some compelling value estimates for people looking to upgrade a whole host of Apple devices.

from the good-deals-on-cool-stuff dept

The Hypergear 3-in-1 Wireless Charging Dock is meticulously engineered to reduce the cable clutter and streamline your daily routine. Featuring 2 dedicated wireless charging surfaces, you can power up your phone and AirPods easily. In addition, you can charge your Apple Watch with the built-in charger mount. Stylish and compact, the dock is perfect for your tabletop, desk, or nightstand and will effortlessly charge your everyday essentials in one convenient place. It’s on sale for $33.

Note: The Techdirt Deals Store is powered and curated by StackCommerce. A portion of all sales from Techdirt Deals helps support Techdirt. The products featured do not reflect endorsements by our editorial team.

Filed Under: daily deal

Your developers are already running OpenClaw at home. Censys tracked the open-source AI agent from roughly 1,000 instances to over 21,000 publicly exposed deployments in under a week. Bitdefender’s GravityZone telemetry, drawn specifically from business environments, confirmed the pattern security leaders feared: employees deploying OpenClaw on corporate machines with single-line install commands, granting autonomous agents shell access, file system privileges, and OAuth tokens to Slack, Gmail, and SharePoint.

CVE-2026-25253, a one-click remote code execution flaw rated CVSS 8.8, lets attackers steal authentication tokens through a single malicious link and achieve full gateway compromise in milliseconds. A separate command injection vulnerability, CVE-2026-25157, allowed arbitrary command execution through the macOS SSH handler. A security analysis of 3,984 skills on the ClawHub marketplace found that 283, about 7.1% of the entire registry, contain critical security flaws that expose sensitive credentials in plaintext. And a separate Bitdefender audit found roughly 17% of skills it analyzed exhibited malicious behavior outright.

The credential exposure extends beyond OpenClaw itself. Wiz researchers discovered that Moltbook, the AI agent social network built on OpenClaw infrastructure, left its entire Supabase database publicly accessible with no Row Level Security enabled. The breach exposed 1.5 million API authentication tokens, 35,000 email addresses, and private messages between agents that contained plaintext OpenAI API keys. A single misconfiguration gave anyone with a browser full read and write access to every agent credential on the platform.

Setup guides say buy a Mac Mini. Security coverage says don’t touch it. Neither gives a security leader a controlled path to evaluation.

And they’re coming fast. OpenAI’s Codex app hit 1 million downloads in its first week. Meta has been spotted testing OpenClaw integration in its AI platform codebase. A startup called ai.com spent $8 million on a Super Bowl ad to promote what turned out to be an OpenClaw wrapper, weeks after the project went viral.

Security leaders need a middle path between ignoring OpenClaw and deploying it on production hardware. Cloudflare’s Moltworker framework provides one: ephemeral containers that isolate the agent, encrypted R2 storage for persistent state, and Zero Trust authentication on the admin interface.

Why testing locally creates the risk it’s supposed to assess

OpenClaw operates with the full privileges of its host user. Shell access. File system read/write. OAuth credentials for every connected service. A compromised agent inherits all of it instantly.

Security researcher Simon Willison, who coined the term “prompt injection,” describes what he calls the “lethal trifecta” for AI agents: private data access, untrusted content exposure, and external communication capabilities combined in a single process. OpenClaw has all three — and by design. Organizational firewalls see HTTP 200. EDR systems are monitoring process behavior, not semantic content.

A prompt injection embedded in a summarized web page or forwarded email can trigger data exfiltration that looks identical to normal user activity. Giskard researchers demonstrated exactly this attack path in January, exploiting shared session context to harvest API keys, environment variables, and credentials across messaging channels.

Making matters worse, the OpenClaw gateway binds to 0.0.0.0:18789 by default, exposing its full API to any network interface. Localhost connections authenticate automatically without credentials. Deploy behind a reverse proxy on the same server, and the proxy collapses the authentication boundary entirely, forwarding external traffic as if it originated locally.

Ephemeral containers change the math

Cloudflare released Moltworker as an open-source reference implementation that decouples the agent’s brain from the execution environment. Instead of running on a machine you’re responsible for, OpenClaw’s logic runs inside a Cloudflare Sandbox, an isolated, ephemeral micro-VM that dies when the task ends.

Four layers make up the architecture. A Cloudflare Worker at the edge handles routing and proxying. The OpenClaw runtime executes inside a sandboxed container running Ubuntu 24.04 with Node.js. R2 object storage handles encrypted persistence across container restarts. Cloudflare Access enforces Zero Trust authentication on every route to the admin interface.

Containment is the security property that matters most. An agent hijacked through prompt injection gets trapped in a temporary container with zero access to your local network or files. The container dies, and the attack surface dies with it. There is nothing persistent to pivot from. No credentials sitting in a ~/.openclaw/ directory on your corporate laptop.

Four steps to a running sandbox

Getting a secure evaluation instance running takes an afternoon. Prior Cloudflare experience is not required.

Step 1: Configure storage and billing.

A Cloudflare account with a Workers Paid plan ($5/month) and an R2 subscription (free tier) covers it. The Workers plan includes access to Sandbox Containers. R2 provides encrypted persistence so conversation history and device pairings survive container restarts. For a pure security evaluation, you can skip R2 and run fully ephemeral. Data disappears on every restart, which may be exactly what you want.

Step 2: Generate tokens and deploy.

Clone the Moltworker repository, install dependencies, and set three secrets: your Anthropic API key, a randomly generated gateway token (openssl rand -hex 32), and optionally a Cloudflare AI Gateway configuration for provider-agnostic model routing. Run npm run deploy. The first request triggers container initialization with a one-to-two-minute cold start.

Step 3: Enable Zero Trust authentication.

This is where the sandbox diverges from every other OpenClaw deployment guide. Configure Cloudflare Access to protect the admin UI and all internal routes. Set your Access team domain and application audience tag as Wrangler secrets. Redeploy. Accessing the agent’s control interface now requires authentication through your identity provider. That single step eliminates the exposed admin panels and token-in-URL leakage that Censys and Shodan scans keep finding across the internet.

Step 4: Connect a test messaging channel.

Start with a burner Telegram account. Set the bot token as a Wrangler secret and redeploy. The agent is reachable through a messaging channel you control, running in an isolated container, with encrypted persistence and authenticated admin access.

Total cost for a 24/7 evaluation instance runs roughly $7 to $10 per month. Compare that to a $599 Mac Mini sitting on your desk with full network access and plaintext credentials in its home directory.

A 30-day stress test before expanding access

Resist the impulse to connect anything real. The first 30 days should run exclusively on throwaway identities.

Create a dedicated Telegram bot, and stand up a test calendar with synthetic data. If email integration matters, spin up a fresh account with no forwarding rules, no contacts, and no ties to corporate infrastructure. The point is watching how the agent handles scheduling, summarization, and web research without exposing data that would matter in a breach.

Pay close attention to credential handling. OpenClaw stores configurations in plaintext Markdown and JSON files by default, the same formats commodity infostealers like RedLine, Lumma, and Vidar have been actively targeting on OpenClaw installations. In the sandbox, that risk stays contained. On a corporate laptop, those plaintext files are sitting ducks for any malware already present on the endpoint.

The sandbox gives you a safe environment to run adversarial tests that are reckless and risky on production hardware, but there are exercises you could try:

Send the agent links to pages containing embedded prompt injection instructions and observe whether it follows them. Giskard’s research showed that agents would silently append attacker-controlled instructions to their own workspace HEARTBEAT.md file and wait for further commands from an external server. That behavior should be reproducible in a sandbox where the consequences are zero.

Grant limited tool access, and watch whether the agent requests or attempts broader permissions. Monitor the container’s outbound connections for traffic to endpoints you didn’t authorize.

Test ClawHub skills before and after installation. OpenClaw recently integrated VirusTotal scanning on the marketplace, and every published skill gets scanned automatically now. Separately, Prompt Security’s ClawSec open-source suite adds drift detection for critical agent files like SOUL.md and checksum verification for skill artifacts, providing a second layer of validation.

Feed the agent contradictory instructions from different channels. Try a calendar invite with hidden directives. Send a Telegram message that attempts to override the system prompt. Document everything. The sandbox exists so these experiments carry no production risk.

Finally, confirm the sandbox boundary holds. Attempt to access resources outside the container. Verify that container termination kills all active connections. Check whether R2 persistence exposes state that should have been ephemeral.

The playbook that outlasts OpenClaw

This exercise produces something more durable than an opinion on one tool. The pattern of isolated execution, tiered integrations, and structured validation before expanding trust becomes your evaluation framework for every agentic AI deployment that follows.

Building evaluation infrastructure now, before the next viral agent ships, means getting ahead of the shadow AI curve instead of documenting the breach it caused. The agentic AI security model you stand up in the next 30 days determines whether your organization captures the productivity gains or becomes the next disclosure.

Apple’s USB-C Magic Mouse is back on sale for about $11 off its usual retail price of $79. At $68, that’s a savings of 14 percent for one of Apple’s best accessories from a company that does not often run sales.

The multi-touch mouse was first released in 2009 with a modest refresh released in 2015 and the addition of a USB-C port in 2024. The rechargeable mouse features gesture controls and automatically pairs with your Mac when connected via USB. The Magic Mouse can also be used with an iPad via Bluetooth, or with a Windows PC, though in that case, functionality would be limited.

Famously, Jony Ive’s design of the Magic Mouse sees its charge port on the underside of the body, rendering it unusable while charging. In 2024 there were rumors of a more comprehensive redesign coming but nothing has materialized since.

Apple

Advertisement

The deal only applies to the white model. 

Follow @EngadgetDeals on X for the latest tech deals and buying advice.

Five years after shutting down facial recognition on Facebook over privacy concerns, Meta is preparing to bring the technology back – this time through its smart glasses. According to reports, the company is developing a feature internally called “Name Tag” that would allow wearers of its Ray-Ban Meta glasses to identify people in real time using facial recognition, with assistance from its built-in AI system.

Meta had previously discontinued facial recognition for photo tagging in 2021, citing the need to find the “right balance” between innovation and privacy. Now, as its wearable ambitions expand, the company appears ready to revisit the technology. The proposed feature would not function as a universal face search engine, but instead would reportedly recognize people connected to users through Meta platforms or those with public profiles.

The move signals a broader shift in how Meta sees AI-powered wearables shaping the future of computing

The company’s smart glasses, developed in partnership with EssilorLuxottica, have become a surprising commercial success, with millions sold last year. Adding facial recognition could differentiate Meta’s hardware as competition intensifies from companies like OpenAI that are developing their own AI-first devices.

However, the plan carries serious privacy and civil liberties implications. Facial recognition has long drawn criticism from advocacy groups concerned about surveillance, misuse, and erosion of public anonymity. Some U.S. cities have restricted law enforcement use of the technology, while lawmakers have raised alarms about its deployment in public spaces. Critics argue that embedding such capabilities into consumer wearables could normalize constant identification in everyday life.

Meta has reportedly debated how and when to release the feature

The company has acknowledged internal concerns about “safety and privacy risks.” The company is also exploring advanced versions of its glasses – internally referred to as “super sensing” – that could continuously run cameras and sensors. In such scenarios, facial recognition would help the AI assistant provide contextual reminders or information based on who the wearer encounters.

For consumers, the technology could offer convenience, especially for accessibility use cases such as helping blind or low-vision individuals identify people nearby. But it also raises questions about consent and transparency. Meta’s current glasses include a visible LED light to signal recording, and discussions are ongoing about how to signal when facial recognition features are active.

What comes next will likely depend on regulatory scrutiny and public response. Meta remains bound by past privacy settlements with regulators, though internal reports suggest some review processes have recently been streamlined. As AI wearables move closer to mainstream adoption, Meta’s approach to facial recognition could become a defining moment in the balance between innovation and personal privacy.

The forthcoming Nintendo Virtual Boy accessory for Switch and Switch 2 can play VR-supported games, . There are four available games to play, including Super Mario Odyssey, Super Smash Bros Ultimate, The Legend of Zelda: Breath of the Wild and Captain Toad: Treasure Tracker.

These aren’t new VR builds of the games, rather they are the versions previously released for the VR set. This was a kit for the original Switch that allowed users to build a cardboard VR headset, among other items.

However, this is very good news for Switch 2 owners as Labo creations generally don’t work with Nintendo’s shiny new console. So this is the only way to experience the VR versions of the aforementioned four games. It’s also worth noting that the Switch 2 upgrade for Breath of the Wild still includes the VR mode.

There are some caveats. The Virtual Boy accessory is available to purchase as a hardware unit or in cardboard. The cardboard version is much cheaper, at $25, and is actually the preferred method for playing these games in VR.

That’s because the hardware version sits on a stand, like the original Virtual Boy, making it harder to move one’s head around. The cardboard headset is free from those constraints. The hardware also includes red filters over the lenses, to better mimic the original experience, but these can be removed.

However, the hardware version is better for playing actual Virtual Boy games, as they were designed for a static headset resting on a table. You’ll have to decide if that trade-off is worth $100. It’s also worth noting that Virtual Boy games , which is a bummer for OG Switch fans.

Both versions of the Virtual Boy accessories , which is the same day several of the retro console’s games head to the Nintendo Switch Online + Expansion Pack subscription service. They can be . We got a chance to try the headset and came away fairly impressed, though noted that the revamped accessory is “just as eccentric and ungainly as the original was three decades ago.”

For those wondering what all the fuss is about, the Virtual Boy was an actual console released by Nintendo all the way back in 1995. It was one of the first mass-market VR devices and, as such, was decades ahead of the curve. It was cumbersome, the games were only in red and there was nothing by way of motion control. Americans only got 14 games before the console was discontinued.

Digital Audio Players, often abbreviated to DAPs, remain one of the most versatile ways to listen to music, offering a self contained alternative to dongle DACs, portable DAC amps, and desktop systems that keep you tethered to a desk. Modern DAPs are no longer just glorified iPods. Today’s models deliver real output power, capable processing, and designs that increasingly resemble mainstream smartphones rather than niche audio gear.

There are solid budget DAPs on the market, but meaningful value tends to thin out quickly as prices drop. That is where the FiiO JM21 becomes interesting. At $179, it lands in a price bracket where most players play it safe, trimming features and performance to hit a number rather than pushing the envelope.

Developed in collaboration with Jade Audio, FiiO’s value focused sub brand, the JM21 does not try to look expensive or pretend it belongs in a higher tier. It is compact, understated, and almost anonymous. Internally, however, FiiO appears to have packed in far more than this category normally allows, from power delivery to functionality and overall flexibility.

That leaves a more uncomfortable question for the competition. Is the JM21 simply good for the money, or did FiiO overdeliver just enough to make nearby alternatives feel needlessly compromised?

Specifications & Technology

At the heart of the JM21 is a dual DAC configuration built around Cirrus Logic CS43198 chips, paired with SGM8262 op amps handling the output stage. That is serious silicon for a budget friendly DAP, and the supporting numbers back it up. FiiO claims a signal to noise ratio of roughly 130dB, total harmonic distortion plus noise below 0.0006%, and support for sampling rates up to 768 kHz at 32-bit, along with DSD512 over USB.

Those figures are not just the result of good parts selection. Internally, the JM21 is laid out with the control section, DAC stage, and amplifier stage physically separated into distinct zones. Each section is further isolated with shielding, a design choice intended to reduce crosstalk and keep noise from creeping into the signal path.

Power delivery is treated with similar care. The JM21 uses a three section power supply, with dedicated regulation for the digital control circuitry, the DAC stage, and the current and voltage amplification stages. The goal is straightforward. Provide stable, uninterrupted power where it matters most, rather than letting everything fight over a single rail. In a player at this price, that level of internal discipline is notable and not something competitors can all claim with a straight face.

Add in extremely low jitter femtosecond crystal oscillators, SRC bypassing, and FiiO’s proprietary DAPS Digital Audio Purification System, and the JM21 starts to look like a player that has been engineered with real intent rather than assembled from leftovers. On paper, the focus is clearly on preserving signal integrity and extracting as much performance as possible from the hardware.

The obvious concern is whether all of this comes at the expense of usability. It does not appear to. The JM21 is powered by a Qualcomm Snapdragon 680 processor, backed by 4 GB of RAM and a customized Android 13 operating system. Performance is responsive, app support is broad, and the interface avoids the lag and stutter that still plague some entry level players.

Internal storage sits at 64 GB, with expansion supported up to 2 TB via a microSD card slot. Battery life is another quiet strength. Thanks to the JM21’s relatively low power consumption, FiiO rates it at up to 12.5 hours of playback, a figure that held up in real world use rather than collapsing the moment Wi-Fi and streaming entered the picture.

Design & Build Quality

Included with the JM21 is everything you need and nothing you do not. In the box you get a transparent plastic case, a basic black USB Type-C to A cable for charging and data transfer, and the usual documentation. No extras, no padding, no pretending this is a luxury experience.

The first thing that stands out when you pick up the JM21 is just how thin and light it is. At 13 mm thick, roughly 0.5 inches, and weighing 156g, about 5.5 ounces, it feels closer to a compact smartphone than a traditional DAP. Its overall dimensions are equally manageable at 120 mm tall and 68 mm wide, or approximately 4.7 by 2.7 inches, making it easy to operate comfortably with one hand.

The chassis is a mix of aluminium alloy and plastic. It does not scream premium, but it feels solid enough to handle everyday use without complaint. The textured underside is a nice touch, adding grip where it actually matters. My review unit was finished in black, though a more eye catching sky blue option is also available. No one is going to confuse this with a flagship build, but at this price point it is sturdy, practical, and frankly hard to fault.

The control layout is straightforward and sensibly arranged. On the left side you will find the power button, which incorporates a small indicator light, along with a volume up and down rocker. Everything falls easily under your thumb, even when using the player one handed.

The right side houses the physical media controls, including play and pause, track forward, and track back buttons. The microSD card slot is also located here, keeping all removable and frequently used controls in one place.

Along the bottom edge are the audio and data connections. The JM21 offers both 4.4mm balanced and 3.5mm unbalanced headphone outputs. The 4.4mm jack can also function as a line out, while the 3.5mm output supports both line out and coaxial digital out. A USB Type-C port rounds things out, handling charging and data transfer duties.

User Experience

Power up the JM21 and you are greeted by a bright, vibrant 4.7-inch TFT touchscreen. While the resolution is a modest 1334×750, it is well matched to the screen size. In practice, text and artwork look clean, and I never found myself distracted by visible pixels.

Beyond its wired outputs, the JM21 also supports wireless listening via Bluetooth 5.0. It can both transmit to and receive from compatible devices, with LDAC support enabling high quality wireless playback at up to 96 kHz. Pairing was quick and stable, and performance was consistent during testing.

Versatility is clearly a priority here. The JM21 can operate in line out mode for use with active speakers or external power amplifiers, and it can also function as a USB DAC. In that configuration, connecting it to a laptop, desktop PC, or even a mobile device is straightforward, allowing the JM21 to bypass inferior onboard audio and handle digital conversion duties itself.

Most of my listening was done in standard Android mode, though FiiO also offers a Pure Music mode for those who want a more focused experience. This mode strips the interface back to the essentials, minimizing background processes and visual clutter so the player behaves more like a traditional, music only DAP. If you prefer fewer distractions and quicker access to your library, it is a sensible option.

Overall performance was stable, but not entirely flawless. I experienced occasional Spotify app crashes, particularly during the first few hours of use. The cause was not immediately clear, though the issue appeared to resolve itself over time and did not persist as testing continued. Outside of that early hiccup, day to day operation was smooth and predictable.

The Android 13 implementation feels familiar and largely hassle free, with no noticeable stuttering or performance limitations during typical day to day use. Navigation is smooth, app switching is responsive, and the overall experience feels appropriately tuned for a dedicated audio device.

I did spend some time with the Pure Music mode, which limits operation to the FiiO Music app. In this configuration, the JM21 behaves like a more traditional DAP, prioritizing local playback and simplicity. The app itself is well executed, offering straightforward access to local files, wireless file transfer to and from a connected phone or computer, and built-in EQ adjustment.

If you maintain a large locally stored music library, Pure Music mode makes a lot of sense. It is faster, cleaner, and avoids the overhead of Android apps you are not using, allowing the JM21 to focus on what it does best.

Listening Impressions and Headphone Synergy

Most of my listening impressions were formed using a mix of Spotify streams and hi res FLAC files stored on a microSD card. I paired the JM21 with a wide range of over ear headphones via the 4.4mm balanced output, including the HiFiMAN HE1000 Unveiled, Sendy Audio Egret, Beyerdynamic DT880 Edition 600 Ohm, and DALI IO-12. The DALI was also used wirelessly to evaluate Bluetooth performance.

In short, the JM21 presents a clean, neutral, and largely uncolored sound signature. It does not impose a strong personality of its own, instead allowing the character of the connected headphones to come through intact. Bass, midrange, and treble are evenly balanced and well integrated, provided the headphones themselves are similarly well tuned. This is not a player that sweetens, exaggerates, or smooths things over. What you hear is largely what your headphones are capable of delivering.

Despite its largely flat, neutral tuning, the JM21 never comes across as sterile or robotic. There is enough body and tonal weight to keep music sounding human rather than processed. On “Feeling Good” by Nina Simone, her vocal carries real heft and authority, sitting front and center with a natural sense of scale. The brass section has proper bite and presence, with trumpets cutting through cleanly and trombones sounding full and weighty rather than thin or splashy. The JM21 keeps these elements in balance, letting the track breathe without smoothing away its character.

That said, the JM21 is not a technical showpiece. Transient snap and large scale dynamics are a bit restrained compared to some similarly priced dongle DACs, which can sound more immediate and energetic in direct comparison. There is a trade off here, however. Those dongles do not give you a full Android experience, onboard storage, or a proper touchscreen interface. Viewed in that context, the JM21’s performance makes more sense. You are trading a bit of outright technical bite for versatility, convenience, and an all in one listening experience that dongles simply cannot offer.

Soundstage and imaging are fairly average, with limited spatial placement. Detail retrieval is solid, however. On “Chocolate Chip Trip” by TOOL, the JM21 still revealed subtle percussive hits and low level effects that many devices gloss over. The issue is scale. The track’s complex spatial placement and sense of movement felt flattened compared to higher quality sources. Everything was audible, but the presentation lacked the depth and dimensionality that make this track truly jaw dropping.

I did not have any similarly priced DAPs on hand, but I did compare the JM21 to the $500 Shanling M3 Plus. In terms of overall detail retrieval and tonal balance, the JM21 more than held its own. The differences came down to refinement. The Shanling sounded slightly more natural in timbre and more convincing dynamically, pulling ahead by a small but noticeable margin rather than a night and day difference.

700 mW is a heck of a lot of power for a sub-$200 device, and that kind of headroom proves useful when driving harder-to-run over-ear headphones like the HE1000 Unveiled. Just do not expect it to unlock the full potential of notoriously demanding models such as the HiFiMAN HE6se V2 or Modhouse Tungsten.

Furthermore, the JM21 doesn’t use tube amplification nor does it contain an R2R DAC, so it won’t do much to tame treble peaks on troublesome headphones. For that reason, the DT880 Edition 600 Ohm was quickly put to one side for the rest of the review process, as it can sound quite harsh on many solid-state devices.

The Bottom Line

The FiiO JM21 is not a giant killer, but it is a smartly engineered reality check. It has ample output power for the money, a clean and neutral sound, excellent versatility with full Android, strong connectivity, and hardware choices that feel deliberate rather than cheap. And it will comfortably drive the vast majority of headphones people actually own.

The trade offs are just as clear. Technical performance is competent rather than exciting, with average dynamics, soundstage, and spatial placement. It will not soften treble heavy headphones, nor will it extract the last ounce of performance from notoriously power hungry or temperamental designs. If you are chasing holographic imaging or tube like warmth, this is not the DAP for you.

Where the JM21 wins is value and usability. At $179, it offers a level of power, functionality, and polish that makes many alternatives feel compromised or redundant. If your priorities are flexibility, sensible tuning, and maximum bang for the buck, the JM21 makes a very strong case that you may not need anything more.

Pros:

• High quality DAC implementation with class leading measurements for the price
• Lightweight, slim design that is easy to operate one handed
• Smooth and familiar Android 13 experience with good overall responsiveness
• Excellent versatility with multiple operating modes, including USB DAC, line out, Bluetooth, and Pure Music mode
• Clean, neutral, and well balanced sound that avoids obvious coloration
• Strong output power for its class, capable of driving most real world headphones

Cons:

• Occasional app instability, particularly with streaming services early on
• Average dynamics, soundstage, and spatial placement compared to more technical sources
• Sonic presentation prioritizes balance and control over excitement
• Limited ability to tame treble heavy or difficult headphone pairings

Where to buy:

Related Reading: