For Linux users, iMessage has always been Apple's most deliberate exclusion — accessible from a Mac, an iPhone, or an iPad, and nowhere else. A developer going by erikwb has quietly changed that with BlueFerry, an open-source project that routes iMessage and SMS from a paired iPhone to the Linux desktop over a direct Bluetooth connection.

No Mac relay. No Apple login. No cloud intermediary. No subscription. Just Bluetooth.

How it works

BlueFerry uses the Bluetooth accessory services already exposed by iOS — specifically the Message Access Profile (MAP), which iOS uses to share messages with accessories like car infotainment systems. Once a Linux machine is paired with the iPhone, BlueFerry intercepts that protocol and presents it through a native desktop interface.

The key detail: iOS decides whether an outgoing message is sent as SMS or iMessage based on the same logic it uses normally — BlueFerry just hands the message to the phone and lets iOS route it. That means iMessage encryption and delivery receipts work as expected for conversations with other Apple users, while SMS falls through for everyone else.

Native clients for GNOME, KDE, and Quickshell

As described on the Linuxiac writeup, BlueFerry ships three desktop clients — one for GNOME, one for KDE, and one for Quickshell — all sharing the same local backend. Pairing data, contacts, message history, and preferences are stored once and remain consistent across whichever client you switch to. Contact sync works out of the box, and the app can optionally mirror other iPhone notifications to the desktop as well.

What it supports, and what it doesn't

For a first release of experimental software, the feature set is broader than expected:

• Receiving and sending iMessage and SMS
• Contact synchronization
• Starting new conversations (by phone number or Apple ID email)
• Optional mirroring of other iPhone notifications

What isn't supported yet: attachments, reactions, typing indicators, FaceTime, voice calls, and a complete sent-message history from before pairing. The developer is candid about the experimental state — development has primarily used an iPhone 16 Pro Max running iOS 26.5, and Apple could change the Bluetooth behavior BlueFerry depends on at any point without notice.

Licensing and getting started

BlueFerry is released under the GPL-2.0-only license. The GitHub repository includes source code, architecture documentation, protocol notes, and testing information for anyone who wants to audit, contribute, or build from scratch.

Installation instructions and desktop environment-specific setup guides are in the repository's README. The project is new enough that distro packages don't exist yet — for now, expect to build from source.

The cautionary note from the developer is worth repeating: don't rely on it as your only path for important messages. But as a first step toward closing one of the longest-standing gaps in the Linux desktop experience, BlueFerry is a genuinely promising start.

AR display glasses have had a persistent, unglamorous problem: nearsighted users either buy custom prescription lenses ($40–$100 extra, weeks of waiting) or squint through a slightly soft image. On August 6, 2026, VITURE released the Pro 2 XR Glasses at $299 — and solved it with built-in, per-eye diopter dials covering 0 to −5.00D. No inserts. No separate order. Twist the dial.

The prescription dial is a bigger deal than it sounds

Birdbath optics — the category VITURE operates in — have a reputation for sharpness degrading toward the edges, and custom prescription inserts typically only help with the center. VITURE's new UltraClarity 3.0 optics were redesigned to correct field curvature across the entire 50-degree field of view, not just the center, and the built-in diopter adjustments work across the full field with them. For the roughly 40% of the population that is nearsighted in the 0 to −5D range, this eliminates the biggest friction point in the category.

Specs that hold up at $299

According to the Road to VR launch coverage, the hardware underneath the price tag is genuinely competitive:

• Display: Sony micro-OLED panels, 1,920 × 1,080 per eye, 120 Hz, 1,600 nits
• Field of view: 50 degrees — up from 46 degrees on the original VITURE Pro
• Virtual screen size: 146-inch equivalent
• Weight: 63g — 20% lighter than the previous model, with a 16% thinner frame
• Eye care: SGS A+ certified, with chromatic aberration corrections and reduced eye strain per VITURE's claims

The 1,600 nits brightness is relevant: most indoor AR displays struggle in moderately lit rooms. Whether the Pro 2 actually holds that figure in practice is something the Tom's Guide review found credible for indoor use, though bright outdoor environments remain a limitation of the birdbath category regardless of manufacturer.

Compatibility is wide

The Pro 2 connects over USB-C and is compatible with Nintendo Switch, Switch 2, Steam Deck, iPhone 15 and later, Android devices, and Mac and PC. VITURE also offers a Mobile Dock Mini for phones that lack native USB-C display output. This plug-and-play approach is deliberately simple — no proprietary wireless protocol, no pairing ceremony, no app required on the source device.

Where it sits in the market

At $299, the VITURE Pro 2 lands at the same price point as competing glasses from RayNeo and Xreal's X sub-brand. What distinguishes it is the prescription dial — neither competitor offers that at this price. The closest alternative is ordering custom inserts separately, which adds cost, delays, and a dependency on fit.

The glasses remain display-only — no on-board compute, no OS, no gesture tracking. If you want a standalone AR computer, this is not it. But if you want a portable large screen that travels in your bag and doesn't require optician appointments, the VITURE Pro 2 is currently the most friction-free way to get one.

More details on VITURE's product page.

On August 11, 2026, the Flatpak project released version 1.18.1 — a critical security update that patches ten CVEs at once, several of which are severe enough to deserve your immediate attention if you run Flatpak applications on Linux.

What Flatpak is and why this matters

Flatpak is the Linux application sandboxing and distribution framework used across Fedora, Ubuntu, Arch, and most other mainstream distributions. It promises that apps run in isolation from your host system — a sandbox. These vulnerabilities shatter that promise entirely.

The worst of them: a full sandbox escape

The most critical fix addresses CVE-2026-34078, a sandbox escape rated CVSS 9.3. The flaw lives in how the Flatpak portal handles paths in the sandbox-expose options. A sandboxed application can craft symlinks pointing anywhere on the host filesystem; when Flatpak mounts the resolved path inside the sandbox, the app suddenly has access to the full host — including system binaries, user home directories, and credential stores. From there, arbitrary code execution in the host context is feasible.

In plain English: any Flatpak app on your system could have read and written anywhere it wanted.

The rest of the ten CVEs

The other nine fixes address a layered set of additional privilege escalation paths, all described in the Linuxiac breakdown:

• Local root privilege escalation via symlink path traversal in revokefs and commit tampering — a direct path to root.
• Arbitrary root file writes in extra-data extraction, exploited through symlinks and path traversal.
• Another arbitrary file write in flatpak build-init using the same mechanism.
• Symlink attack on .ld.so, allowing files with arbitrary names to be written anywhere on the filesystem.
• A downgrade loophole that let unprivileged users bypass safeguards and force-install older, potentially vulnerable app versions.

Concurrent releases: 1.19.0 pre-release also patched

Flatpak simultaneously pushed 1.19.0 (a pre-release branch) with the same security fixes, ensuring users on both the stable and development branches are covered. This simultaneous coordinated release is a signal of how serious the project considered these issues.

What you should do now

Update immediately. On most distributions, sudo apt upgrade flatpak, sudo dnf upgrade flatpak, or the equivalent will get you there. You can verify your installed version with flatpak --version — you want 1.18.1 or later.

If you use Flatpak in production or in any multi-user environment, treat this as an emergency patch. The scope of the vulnerabilities — full host filesystem access from any installed application — is about as bad as sandboxing failures get.

Flatpak's maintainers deserve credit for shipping a coordinated, comprehensive release rather than drip-feeding individual patches. Ten CVEs in one shot is a lot, but it is far better than discovering them one by one in the wild.

Yesterday, Obsidian Entertainment shipped the "Into the Abyss" update for Grounded 2 — and at the same time, the game arrived on PlayStation 5 for the first time. That's a lot happening at once, and both parts are genuinely significant.

What "Into the Abyss" Actually Adds

The update opens the pond in Brookhollow Park, a biome that expands the game map considerably both above and below the waterline. This isn't a cosmetic reskin — the update introduces underwater base building, a new class of creatures, and a craftable diving suit that extends how long you can stay submerged before you need to surface. The deeper you go, the more dangerous and rewarding the pond becomes.

Two additional biomes also land with this update: the Mountains and the Bayou, each with their own discovery loops, hazards, and loot. The Toe Biter Buggy — an amphibious vehicle that handles both land and water — gives players a way to navigate the expanded geography without constantly swapping gear.

For context, Grounded 2 entered Early Access on PC and Xbox Series X/S in July 2025. This is the largest content drop since that launch, and Obsidian is packaging the entire thing as a free update for anyone who already owns the game or accesses it through Xbox Game Pass.

The PS5 Launch

The PlayStation 5 release is notable for a couple of reasons. First, Grounded 2 started as an Xbox-exclusive Early Access title, so bringing it to Sony's platform is a continuation of Microsoft's broader shift toward multiplatform publishing. Second, PS5 players are jumping in at the same moment as the biggest update to date — they're not inheriting a stale build.

Full cross-play support between PS5, Xbox Series X/S, and PC is included, so the player pool doesn't fragment across platforms. PS5 buyers are also getting the Early Access version, not a 1.0 launch — Obsidian hasn't announced a full release date yet.

Where Things Stand

The original Grounded launched in Early Access and spent roughly a year building out its content before its 1.0. Grounded 2 appears to be following a similar cadence — big, themed updates every few months, each one expanding the map and mechanics meaningfully. "Into the Abyss" is the water chapter. What comes next is still open.

If you're on any platform and haven't revisited the game in a while, this is a good moment to check back in. The pond alone adds a substantial new layer to the survival loop, and underwater base building is a mechanic that changes how you think about resource management entirely.

CachyOS — the performance-focused Arch-based Linux distribution — shipped its August 2026 release this week, and the headline change is a complete rewrite of its default graphical package manager, Shelly. The new version drops C# entirely in favor of Zig.

Why Rewrite Shelly?

Shelly was originally built in C#, which meant shipping a managed runtime alongside it. The runtime added overhead — longer startup times, higher idle memory use — that didn't sit well with a distribution built around raw performance. The rewrite to Zig produces native binaries with no runtime dependency, which translates to noticeably faster launch and a smaller memory footprint.

The new Shelly isn't just a faster port of the old one. The desktop interface was redesigned: a first-run welcome screen, list and grid package browsing, AUR PKGBUILD previews with visible build output, and a Utilities page for database sync, cache cleanup, and orphan package removal. The CLI side got a full rebuild as well — integrated repository and AUR search, combined update checks covering repositories, AUR, AppImages, and Flatpaks, plus TOML-based config backup and restore.

Flatpak support is now an optional add-on via the shelly-flatpak-backend package rather than baked in — a clean separation for users who don't want Flatpak in their workflow.

Rust Is Also Taking Over

The system tray update applet, Cachy-Update, was rewritten in Rust for the same reasons: native compilation, low overhead, and better reliability. CachyOS is incrementally moving its own tooling away from managed runtimes, using Zig for the package manager and Rust for system utilities. It's a coherent strategy: both languages give you memory safety and native performance without a garbage collector.

Server Edition Is Coming

The most forward-looking part of the August release is quieter: the CLI installer now includes experimental installation profiles for an upcoming CachyOS Server Edition. Nothing has launched yet, but the groundwork is being laid in the distribution's tooling. CachyOS's kernel patches — which include optimizations for latency, CPU scheduler tuning, and throughput — are well-suited to server workloads, so a server-targeted release has been a natural extension for a while.

The Bigger Picture

CachyOS has been methodically improving its out-of-the-box experience over the past year. The August release also lands KDE Plasma 6.7.4, Linux 6.18 LTS and 7.1 kernels, and updated Hyprland dotfiles — the usual cadence of a rolling release. But the Shelly rewrite is the kind of change that reshapes the day-to-day feel of the distribution. Package management is one of the first things you touch in any Arch-based system, and a faster, cleaner tool matters more than most individual package updates.

If you've been running CachyOS, the Zig-based Shelly should land in the next system update automatically. If you've been curious about it as a daily driver, this is probably the most polished the distribution has felt.

At the Future of Memory and Storage (FMS) 2026 conference in Santa Clara last week, Samsung unveiled three new memory and storage architectures aimed squarely at the AI infrastructure bottleneck. The most striking of the three — zHBM — rethinks where memory lives in a system entirely.

The Core Problem zHBM Solves

High Bandwidth Memory (HBM) is the memory type used in today's AI accelerators: the H100, the MI300X, and their successors. The standard arrangement places HBM beside the GPU die, connected via a silicon interposer. That works, but the interposer consumes physical space and imposes bandwidth limits. As AI models grow, those limits are becoming a genuine constraint on what hardware can do.

zHBM takes a different approach: it stacks memory directly above the AI accelerator, using advanced wafer bonding to integrate the two dies vertically. Samsung projects zHBM will deliver approximately 8× the performance of HBM5 and more than 10× the memory density. Those aren't incremental numbers. If the projections hold through actual productization, zHBM represents a step-change in what an AI chip can do with local memory.

The Other Two Technologies

Samsung also introduced zNAND-O, a next-generation NAND architecture built for edge AI environments. It comes in four- and eight-layer configurations, targeting real-time, data-intensive applications where latency matters as much as throughput. Edge AI inference — running models on devices rather than in data centers — is an increasingly important market, and zNAND-O is aimed directly at it.

The third announcement was V10 BV-NAND, the industry's first 400-plus-layer NAND enabled by wafer bonding. The wafer bonding technique separates the array and peripheral circuits onto different wafers before bonding them together, which is what makes 400+ layers achievable. V10 BV-NAND increases storage density by roughly 58% over Samsung's previous V9 generation.

All three technologies share the same foundational ingredient: wafer bonding. It's a manufacturing technique that joins two separately fabricated wafers at the die level, and it's what lets Samsung push density and performance in ways that traditional stacking approaches can't.

When Does This Ship?

Samsung positioned all three as roadmap items, not products you can order today. zHBM and zNAND-O are concept architectures; V10 BV-NAND is further along but still pre-production. The FMS presentation laid out the memory roadmap through HBM4E and HBM5, situating these technologies in the longer arc of AI infrastructure scaling.

What matters now is that the architectural direction is public and specific. Competitors are watching, and the wafer bonding approach will likely force responses across the industry. The memory wall — the gap between processor speed and memory bandwidth — has been the defining hardware problem of the AI era. Samsung just showed one direction for breaching it.

In early August 2026, OpenAI published something unusual: a 249-page mathematical manuscript containing proofs for ten open problems — each one unsolved for at least a decade, some for much longer. The work was produced by an internal build of Astra, OpenAI's next major model. Total compute cost: approximately $2,000.

The number that makes this story strange isn't the $2,000. It's the ten.

What Was Actually Solved

The problems span high-dimensional geometry, group theory, quantum complexity, lattice cryptography, and extremal combinatorics. OpenAI hasn't released the full list with individual attribution yet, but the most significant result — and the one that has mathematicians paying closest attention — is a proof that non-sofic groups exist.

Soficity is a property related to how well a group can be approximated by finite symmetric groups. The question of whether every group is sofic was posed by mathematician Mikhail Gromov in 1999 and became one of the more persistent open questions in geometric group theory. Astra's proof, if it holds up to peer review, would resolve it negatively: there exist groups that are genuinely non-sofic, meaning the approximation fails in a fundamental way.

Machine-Checkable Proofs

OpenAI published the manuscript alongside machine-checkable Lean 4 certificates on GitHub. Lean 4 is a formal proof verification system — feeding a proof into Lean either confirms it is logically valid or identifies exactly where it breaks down, with no room for human error in the checking process. This is significant: it means any mathematician with a computer can verify the proofs themselves in minutes, without having to trust OpenAI's claims or wait years for traditional peer review.

The ability to generate trustless, instantly-verifiable mathematical proofs is arguably more important than any individual result. It changes the economics of mathematical verification entirely.

Why $2,000 Matters

The cost figure is deliberately highlighted by OpenAI, and for good reason. Mathematical research at the frontier typically requires years of expert human effort. A Fields Medal winner might spend a decade on a single hard problem. The cost per solved problem here — roughly $200 — isn't just cheap. It reframes the question of what "doing mathematics" looks like at scale.

If Astra can reliably solve decade-old open problems at this cost, the implication isn't just that AI is good at math. It's that the backlog of unsolved mathematical problems — there are thousands of them, at varying difficulty levels — could begin to clear at a pace that would have seemed absurd two years ago.

What Comes Next

OpenAI hasn't announced a public release date for Astra. The math manuscript appears to be a deliberate capability preview, similar to how DeepMind used AlphaFold protein structure results before the full system became available. The Lean 4 certificates are live and human mathematicians are already working through them.

The non-sofic groups proof in particular is drawing attention. Several group theorists contacted by Forbes described the result as credible pending review, while noting that the Lean verification provides a level of assurance that doesn't exist for most claimed proofs. Whether every result survives close scrutiny remains to be seen — but the bar for dismissal is now much higher when the proof checker already said yes.

Sources: The Decoder — Astra math announcement · Forbes — $2,000 and ten problems

If you run Linux — on a server, a desktop, or inside a container — there is a high-severity kernel vulnerability you need to patch right now. CVE-2026-64564, nicknamed SCTPhantom by researchers at Tencent's Zhuque Lab, was disclosed on August 6–7, 2026. The bug has been sitting in the Linux kernel since December 2007 — nearly 18 years — undetected.

What the Bug Does

SCTPhantom lives in the SCTP (Stream Control Transmission Protocol) subsystem, specifically in the Dynamic Address Reconfiguration (ASCONF) code path. It is a use-after-free vulnerability: memory that has already been freed is accessed again, and under the right conditions an attacker can turn that into arbitrary code execution with kernel privileges.

The practical impact is severe. An unprivileged local user can escalate to full root on the host. More critically for cloud and containerized environments, the exploit enables container escape — an attacker inside a Docker or Kubernetes container can break out to compromise the underlying host machine. In a shared hosting or multi-tenant environment, that means one compromised container can threaten everything else running on the same machine.

Who Is Affected

The vulnerable code was introduced in Linux kernel 2.6.25. Any distribution shipping a kernel derived from that version — which is essentially all of them — is affected unless patched. Confirmed vulnerable distributions include:

• Debian 13
• Ubuntu 24.04 LTS
• Rocky Linux 9
• Red Hat Enterprise Linux 9
• OpenCloudOS

If your kernel is older than the patched versions listed below, assume you are vulnerable.

Patches Are Available — Update Now

Fixes have been backported to all active stable branches. The patched kernel versions are:

• 6.6.148 (LTS, supported through December 2027)
• 6.12.101 (LTS, supported through December 2028)
• 6.18.42 (LTS, supported through December 2028)
• 7.1.6 (mainline)

Notably, new LTS stable kernels (6.18.44, 6.12.103, 6.6.151) were pushed on August 9th — if you've already updated to those, you're covered. For Ubuntu users, running sudo apt update && sudo apt upgrade and rebooting will pull the fix automatically once Canonical has pushed it to their repositories.

Why Did It Take 18 Years?

The SCTP protocol is relatively niche — most traffic runs over TCP or UDP — which means the affected code path sees less scrutiny than the network stack's more heavily trafficked corners. Use-after-free bugs can also be notoriously hard to trigger deterministically; they often require precise timing and memory layout, making them easy to miss in automated fuzz testing unless the fuzzer is specifically targeting that subsystem.

Tencent's Zhuque Lab hasn't disclosed exactly how they found the bug, but the disclosure follows the responsible disclosure timeline — patches were prepared before public announcement, and no in-the-wild exploitation has been confirmed as of the disclosure date.

Workaround If You Can't Patch Immediately

If an immediate kernel update isn't possible, you can disable the SCTP module:

echo "install sctp /bin/true" | sudo tee /etc/modprobe.d/disable-sctp.conf
sudo modprobe -r sctp

This blocks SCTP entirely and mitigates the vulnerability without a reboot, though it will break any services that rely on SCTP (rare for most deployments, but check before applying). The proper fix remains patching to a fixed kernel version as soon as your distribution makes one available.

Sources: The Hacker News — SCTPhantom disclosure · Linuxiac — vulnerability details · Privacy Guides — full system takeover

On August 10, 2026, Meta quietly dropped what may be the most significant open-source AI release of the year: Muse Glimmer, a 30-billion-parameter dense multimodal model published under the Apache 2.0 license. Unlike most frontier models that live behind API paywalls, Muse Glimmer is designed to run on personal hardware — your workstation, your laptop, your local server.

What Makes Muse Glimmer Different

The model punches well above what its parameter count might suggest. It ships with a 131,000-token context window, supports over 100 languages, and is specifically tuned for three high-value use cases: agentic tool use, code generation, and serving as an LLM-as-judge in automated evaluation pipelines.

For developers, that last capability is particularly useful. Rather than paying API costs to evaluate model outputs at scale, Muse Glimmer can act as a local evaluator — checking correctness, consistency, and quality of other models' outputs without sending data off-device. At 30B parameters, it can fit on a single high-end consumer GPU or a modest multi-GPU workstation rig.

Apache 2.0: The License That Actually Matters

Meta chose Apache 2.0, not a custom "open" license with commercial restrictions. This means developers and companies can use, modify, fine-tune, and redistribute Muse Glimmer — commercially — without negotiating special terms. It's a meaningful distinction in a field where "open source" often means "free until you make money from it."

The release puts competitive pressure on proprietary model providers. With a state-of-the-art 30B model available locally under a permissive license, the argument for sending sensitive data to a third-party API weakens considerably.

Zuckerberg's AI Concentration Warning

CEO Mark Zuckerberg accompanied the release with an unusually direct statement about AI governance. He warned publicly about the risks of allowing a small number of companies or governments to gain exclusive control over the most capable AI systems — a pointed message in the same week that new EU and California AI regulations came into force.

The statement reads less like corporate positioning and more like a genuine ideological argument: that concentrated AI power, regardless of who holds it, is a structural risk. Meta's bet is that a world with widely accessible open-source frontier models is safer than one where capability is locked behind a handful of gatekeepers.

What It Means for the Ecosystem

Muse Glimmer arrives alongside Meta's coding agent Muse Code, which launched five days earlier and explicitly targets Claude Code and OpenAI's Codex. Together, they signal that Meta's 2026 AI strategy is to win developers with openness: permissive licenses, local deployment, and aggressive pricing for those who prefer the cloud.

For teams building AI-augmented workflows, privacy-conscious organizations, or anyone who's been waiting for a genuinely capable local model, Muse Glimmer is worth a serious look. The weights are available on the Meta Llama Hub on Hugging Face.

Sources: TechCrunch — Meta launches Muse Code · CNBC — Meta debuts Muse Code · AI Weekly — August 10, 2026

Valve's long-awaited standalone VR headset has a price problem. According to insider Tyler McVicker, the Steam Frame will start at "well over $1,100" — a significant jump from Valve's original sub-$999 target. The report landed on August 2, reigniting debate about whether the headset can succeed at a premium price point in a market where Meta's Quest 3 sells for $499.

The culprit is familiar: supply chain pressure. Memory chip shortages and rising costs for the Snapdragon 8 Gen 3 — the processor powering the Steam Frame — pushed manufacturing costs up enough that Valve couldn't hold its original price. McVicker, who has a reliable track record on Valve hardware, gave no firm number beyond "well over $1,100."

Despite the unwelcome figure, every other signal points to a launch that is extremely close. The Steam Frame has cleared 22 known customs shipments, completed its FCC filing, and has an active store merchandising page live on Steam — all indicators that typically precede an imminent retail release. McVicker described the window as "weeks away."

The hardware itself remains genuinely compelling. The Steam Frame runs SteamOS as a standalone headset — no PC required — with a Snapdragon 8 Gen 3, 16GB of RAM, and dual 2160×2160 displays. It's a streaming-first design, meaning it prioritizes PC VR streaming over purely local standalone play. That positions it squarely at players who already have a gaming PC and a large Steam library, rather than the casual Quest audience.

The ecosystem is already taking shape. The community-made Half-Life 2: VR Mod recently achieved Steam Frame Verified certification — one of the first titles to do so before the hardware has even launched, enabled by eye-tracked foveated rendering. With Valve controlling both the platform and the certification process, the verified library will grow quickly post-launch.

At $1,100+, the Steam Frame sits above the Meta Quest 3 but below the Apple Vision Pro's original launch price. It's a different product than the Quest — higher-fidelity displays, native SteamOS, and access to the full PC VR catalogue — but the value proposition hinges entirely on how deeply invested a buyer is in the Steam ecosystem. For that audience, it may be exactly what they've been waiting for.

Expect an official announcement soon.

Sources: Notebookcheck · VR.org · UploadVR

After a limited early-access window that opened on July 23, Black Forest Labs has made FLUX 3 Video publicly available as of August 5, 2026. The model generates clips up to 20 seconds long at native HD or Full HD resolution — with audio baked directly in.

That last part is the biggest differentiator. FLUX 3 Video includes dialogue, sound effects, and ambient noise across 14+ languages, generated in a single pass without needing a separate audio model layered on top. Competitors like Sora and Runway Gen-4 have taken different approaches to audio integration, but BFL is positioning bundled audio as a core feature rather than a bolt-on.

The generation modes cover the full creative range: text-to-video, image-to-video, keyframe-to-video, video continuation, and multi-scene camera moves. That last mode is especially notable — it lets creators string together distinct camera setups within a single generation, instead of manually splicing clips together in post-production.

On the benchmarks, BFL is claiming the top Elo scores across both text-to-video (1,135) and image-to-video (1,051) on the Artificial Analysis leaderboard — putting it ahead of Seedance 2.0 and Gemini Omni Flash. Independent validation is still catching up, but the early-access window gave a broad enough pool of creators to make the claims credible.

Pricing is tiered by quality and generation mode. Draft HD starts at $0.06 per second, while full-quality Full HD video-to-video reaches $0.53 per second. For a 20-second full-quality clip, that works out to roughly $10.60 — premium pricing, but in line with what comparable frontier video models charge.

Black Forest Labs built its reputation on the FLUX image generation models, which became a staple of the open-source creative community after their 2024 release. FLUX 3 Video is a deliberate push into the increasingly crowded text-to-video space, where ByteDance (Seedance), Google (Veo), OpenAI (Sora), and Runway are all competing aggressively on quality, speed, and price.

The removal of the early-access waitlist is a meaningful signal: BFL believes the model is ready for sustained production workloads. If the leaderboard claims hold up under real-world use, FLUX 3 Video will attract serious attention.

Sources: The Decoder · Black Forest Labs Blog

The wait for the third and final chapter of the Final Fantasy VII Remake trilogy just got a little more bearable. Square Enix has confirmed that Final Fantasy VII Revelation will appear at Gamescom Opening Night Live on August 25, 2026, with game director Naoki Hamaguchi on stage to deliver what the studio calls a "new look" at the game.

The announcement landed on August 7–8, weeks before the show itself — a deliberate tease that sent the FF7 community into a predictable frenzy. Gamescom Opening Night Live, which airs at 2PM EST on August 25, has become one of gaming's most-watched pre-show events, regularly drawing seven-figure viewership for new reveals.

What do we know so far? Final Fantasy VII Revelation is the conclusion to a trilogy that started with FF7 Remake in 2020 and continued with FF7 Rebirth in 2024. The title targets Spring 2027 for release on PS5, PC, Nintendo Switch 2, and Xbox Series X|S — a simultaneous multiplatform launch that marks a clean break from the PlayStation exclusivity windows that defined the first two installments.

The Spring 2027 window carries additional weight: it lands on the 30th anniversary of the original Final Fantasy VII, which released in Japan on January 31, 1997. Square Enix has a history of leaning hard into anniversaries, and the timing feels deliberate.

The direction of the third chapter remains closely guarded. FF7 Rebirth ended on a devastating, interpretively ambiguous note — multiple timelines, fate branches colliding, and a climax that left the community parsing individual frames for months. Revelation will need to resolve all of it while delivering an ending worthy of one of gaming's most beloved stories.

Hamaguchi's Gamescom appearance isn't just a trailer drop; he's listed for a full presentation, which suggests at minimum a gameplay breakdown and likely some story context. The community's attention will be laser-focused on what Revelation's opening chapters look like, how the combat system has evolved from Rebirth, and whether any of the trilogy's lingering narrative threads are finally addressed.

Mark August 25 on your calendar.

Sources: Nintendo Life · My Nintendo News

Every developer who has pushed Three.js into a complex scene has hit the same ceiling: lights are expensive. Traditional forward rendering evaluates every light against every fragment, so adding more lights multiplies the cost. The usual workaround — a maximum of 8–16 dynamic lights in most scenes — has been a silent constraint on what's feasible in browser 3D for years.

Three.js r185 breaks that ceiling. The release ships ClusteredLighting, a forward+ shading system that partitions the view frustum into a 3D grid of "clusters" — tiles of screen space combined with depth slices. At render time, only lights that intersect a given cluster are evaluated for pixels inside it. Community benchmarks from the beta show approximately 240 fps vs 50 fps with 1,024 point lights on an RTX 4090 running the WebGPU renderer, and a measured 1,588% throughput improvement at 16,384 lights. That's a different category of scene complexity than what was previously feasible.

ClusteredLighting lives on the WebGPURenderer path and replaces the older TiledLighting addon, which has been removed. Migration is a drop-in swap — add the node to your post-processing chain and the renderer handles the rest.

The Nanite rasterizer

The second major addition in r185 is architecturally bolder: a GPU-driven compute rasterizer inspired by Unreal Engine's Nanite LOD system. Rather than issuing traditional draw calls that funnel geometry through the fixed-function rasterizer, this path submits geometry processing as compute shader workloads — enabling GPU-driven culling, LOD selection, and rasterization without CPU round-trips.

Two new examples ship with it: a base version and a variant with image-based lighting. It's experimental territory, but the direction is clear — the Three.js team is building toward scenes where the CPU is almost entirely removed from the per-frame draw decision loop.

WebGPU maturation

r185 is the release where the WebGPU renderer crosses from "interesting experiment" to "handles real-world scenes." Several long-standing gaps are closed:

• WebXR + WebGPU now work together — immersive sessions run on the modern renderer path for the first time.
• ExternalTexture fully supported — live <video> elements and camera feeds can be used as textures on WebGPU without fallback.
• textureGather / textureGatherCompare — TSL now supports these WGSL gather operations, enabling efficient PCF soft shadows in a single sample.
• storageTexture3D — new node type for 3D storage textures, useful for volumetric effects and compute-driven rendering.
• WGSL diagnostic messages — GPU shader errors now surface meaningful diagnostics in the console instead of opaque failures.

Breaking changes worth knowing

r185 also does housecleaning. Several deprecated APIs are removed:

• TiledLighting is gone — use ClusteredLighting.
• AnamorphicNode is removed — use BloomNode.
• positionLocal no longer mutates internal vertex transformations; use positionGeometry instead.
• TSL function renames: directionToColor() → packNormalToRGB(), colorToDirection() → unpackRGBToNormal().
• DRACOLoader and DRACOExporter no longer bundle encoders/decoders — switch to CDN loading.

The r184 → r185 migration guide is comprehensive. For most projects not using the removed addons, the upgrade is straightforward.

The bigger picture

Taken together — ClusteredLighting, the compute rasterizer, WebXR/WebGPU parity, and the ongoing TSL API stabilization — r185 reads as the release where the library's WebGPU bet starts paying off. The browser is getting a GPU API capable of competitive real-time rendering, and Three.js is increasingly able to use the full depth of it. For anyone building in-browser 3D today, this is the version to be on.

Source: Three.js r185 GitHub Release · Three.js Changelog

Ubuntu point releases follow a predictable rhythm: the original LTS ships in April, and the first point release — which bundles all critical bug fixes and, crucially, opens the upgrade path from the previous LTS — arrives around four months later. Right on schedule, Ubuntu 26.04.1 LTS landed on August 4, 2026, marking the moment that "Resolute Raccoon" transitions from early adopter territory to a stable upgrade destination for the wider Ubuntu ecosystem.

If you've been sitting on Ubuntu 24.04 LTS waiting for the green light, that light is now green.

What's in the point release

26.04.1 is not a feature release — it's a quality release. The ISO images now carry four months of accumulated bug fixes, security patches, and hardware enablement updates that would otherwise require downloading the original April ISO and immediately running a lengthy apt upgrade. For new installs, this matters: fresh deployments start from a stable baseline instead of a day-one snapshot with hundreds of pending patches.

More importantly, the point release is the official trigger for upgrade-manager to offer the 24.04 → 26.04 path to existing LTS users. Canonical deliberately withholds this prompt until the first point release to ensure the upgrade experience is smooth — a policy born from past transitions where eager early upgrades hit rough edges in the first weeks of a new LTS.

What 26.04 LTS brings to the table

For users coming from 24.04, the underlying platform changes in "Resolute Raccoon" are substantial. The full release notes cover the headline additions:

• GNOME 50 — the latest major release of the GNOME desktop, with redesigned system settings and improved multi-monitor handling.
• Linux kernel 7.x — broader hardware support, improved energy efficiency on modern CPUs, and continued Wayland driver improvements.
• Wayland by default everywhere — the Xorg session remains available, but Wayland is now the default for all supported GPU drivers including proprietary NVIDIA.
• TPM-backed full-disk encryption — new installs can now use TPM 2.0 for keyless FDE, enabling automatic unlock at boot without a passphrase on trusted hardware.
• Rust in the core — several low-level system components have been replaced or supplemented with Rust implementations, improving memory safety at the OS layer.

How to upgrade

If you're on Ubuntu 24.04 LTS with updates enabled, you'll see an upgrade prompt appear via the Software Updater GUI within a few days of 26.04.1's release. You can also trigger it immediately from the terminal:

sudo do-release-upgrade

The upgrade path is tested and supported. If you're running a server or a production system, the usual advice applies: snapshot first, read the release notes for deprecated packages in your stack, and run a test upgrade in a staging environment before the real thing.

For fresh installs, the 26.04.1 ISO is available now from ubuntu.com/download and is the recommended starting point for any new Ubuntu 26.04 deployment going forward.

Source: Ubuntu 26.04 LTS Official Release Notes

If you've ever watched your laptop fan spin up to ludicrous speed while running next dev, Next.js 16.3 was built for you. Released August 3rd, this update is the biggest quality-of-life improvement the framework has shipped since version 16.0 landed last November.

The headline number is hard to ignore: up to 90% less RAM during development sessions. Next.js dev servers have always been memory-hungry — the Turbopack compiler, the file-watcher, and hot-module replacement all accumulate allocations, and on large codebases it wasn't unusual to hit Node's 4 GB heap limit mid-session. The Register described the fix as "aiming to reduce the dreaded fatal error messages" — and that's exactly what it does. Vercel's solution combines leaner internal data structures with aggressive incremental compilation: only modules that actually changed since the last save get reprocessed, while everything else is served from a persistent on-disk cache.

The same caching principle applies to production builds. Repeat next build runs now detect unchanged artifacts and skip their recompilation entirely — a change that noticeably shrinks CI build times on large monorepos.

Instant Navigations

The second headliner is Instant Navigations: client-side transitions that feel SPA-fast while preserving all of Next.js's server-driven semantics. Previously, navigating between two RSC-heavy pages involved a network round-trip for each segment's server payload. In 16.3, the router speculatively prefetches and caches those payloads in the background, so by the time a user clicks a link the data is already local. The result is imperceptibly fast transitions — without sacrificing progressive enhancement or the SEO benefits of server rendering.

This is a meaningful distinction from going all-in on a client-side SPA. With Instant Navigations, you still get full streaming, React Server Components, and zero-JS route segments where appropriate — you just don't pay a latency cost for navigating between them.

TypeScript 7 and AI-native tooling

16.3 also ships first-class support for TypeScript 7's type-checker, which Vercel says cuts next build type-checking time by 40–60% on typical projects. TypeScript 7 rewrites the type-checker in Go, bringing dramatic compilation speed gains, and Next.js 16.3 is one of the first major frameworks to integrate it natively.

On the AI tooling front, the framework now generates versioned documentation snapshots. AI coding agents (Copilot, Cursor, Claude Code) can automatically pull docs matched to the exact installed version of Next.js, rather than relying on potentially stale training data. It's a small change with outsized ergonomic impact: agent-generated code suggestions will reference APIs that actually exist in your installed version.

Backward compatibility

No breaking changes. Upgrading from any 16.x release requires only a version bump in package.json. The next.config format and page-router conventions are unchanged; App Router users get all the benefits transparently.

For a framework powering a significant share of React production applications, 16.3 reads less like a feature release and more like a reliability milestone — the kind of update that makes daily development feel like the tool is working with you, not against you. If you've been putting off that upgrade, the 90% memory reduction alone makes it worth doing today.

Source: Next.js Blog — Next.js 16.3