Git 2.55 is out, with contributions from more than 100 developers, 33 of them new to the project. The version makes one architectural commitment that will ripple through every Linux distribution and CI pipeline that builds Git from source: the Rust compiler is now required by default. You can still opt out at build time, but Rust is no longer optional. For one of the most foundational tools in software development, this is a meaningful shift.

Rust by Default — What Changes for You

Git has been integrating Rust incrementally since version 2.54, targeting memory-safety improvements in performance-critical paths. Making the Rust compiler a build dependency by default signals that the Git project considers this integration stable enough for every distro to ship. In practice, package maintainers for Debian, Fedora, Arch, and others will need to add a Rust toolchain to their Git build pipelines if they haven't already. For most developers who install Git via a package manager, nothing changes — distro maintainers will handle it. It matters most if you build Git from source yourself.

Parallel Hook Execution

Git 2.54 introduced config-based hooks — defining pre-commit, post-merge, and other hooks in your .gitconfig rather than only as executable files dropped in $GIT_DIR/hooks. Git 2.55 extends that by allowing compatible hooks to run in parallel. A project with independent pre-commit hooks for linting and unit tests can declare hook.<n>.parallel = true for each and Git will run them concurrently. Parallelism depth is configurable globally via hook.jobs, per hook event via hook.<event>.jobs, or inline with git hook run -j.

In monorepos with expensive pre-commit pipelines — running a type-checker, a linter, and a test suite before every commit — this can shave meaningful seconds off each cycle without changing anything about the checks themselves.

git history fixup

Git 2.54 introduced the experimental git history command. Version 2.55 adds a fixup subcommand: stage your changes, run git history fixup, and Git applies the staged diff to an earlier commit in your branch, then replays subsequent commits on top. It's a first-class version of the git rebase -i + fixup! commit workflow that experienced Git users already rely on — just without the manual dance of writing a fixup commit, launching the interactive rebase editor, and reordering lines.

Large-Repo Performance Gains

Git 2.55 expands the --path-walk flag — a packing optimization that groups objects by path for more efficient traversal — to work with blob:none, blob:limit=, tree:0, object:type=, sparse, and combine filters. This unlocks path-walk packing in more partial-clone and filtered-pack workflows, which translates to measurable speedups when cloning or fetching repositories the size of the Linux kernel tree. Benchmarks cited in the release notes show significant time reductions on large fetch operations in repositories with complex sparse-checkout configurations.

Git 2.55 is available now from the official Git releases page and is rolling out to major Linux distribution repositories.

After 18 months in Early Access, Pocketpair has confirmed that Palworld 1.0 launches on July 10, 2026 — simultaneously on PC (Steam), PlayStation 5, Xbox Series X|S, and Game Pass. Anyone who already owns the game gets the update free. No save wipes: your Pals, bases, and progress carry over intact, though Pocketpair recommends a fresh start to experience the new content in sequence.

What 1.0 Actually Adds

The patch notes run to 27 PDF pages. Pocketpair's global communications head admitted he was "losing his mind" trying to format an update this large, and the studio called it "bigger in scale than any update we have ever released." The headliners:

• Wing Pack — aerial traversal gear that lets players fly freely across the world
• World Tree Zone — a dedicated endgame region with new bosses, items, and a narrative conclusion
• Sky Islands — a second major island landmass that roughly doubles the playable area from the original 2024 launch
• PvP Mode — proper competitive multiplayer, one of the most-requested features since the game released
• Genetic Recombination breeding — a deeper Pal genetics system for players who want to optimize their team

The two new regions — World Tree and Sky Islands — were built assuming players discover them progressively, which is why Pocketpair is nudging returning players toward a new save even though old saves are technically fine.

From Overnight Sensation to Full Release

Palworld's Early Access debut in January 2024 was one of the most chaotic launches in recent gaming history: 8 million copies in the first week, servers collapsing under load, and instant cultural saturation. What followed was equally messy — a Nintendo lawsuit over Pal character designs, debates about whether Pocketpair used AI-generated assets, and the usual Early Access skepticism about whether content would ever actually finish shipping.

The 1.0 launch is the answer. The scope of the update — doubled map, core systems like PvP and flight, and a proper narrative ending — makes it a content completion rather than a quiet "we're calling it done" formality. Whether the lawsuit is resolved by then remains separate from what players will actually get on July 10.

Should You Return?

If you played at launch and dropped off, July 10 is a reasonable reentry point. The Wing Pack alone changes traversal fundamentally, the Sky Islands are designed to feel like a proper second act, and PvP gives multiplayer servers a competitive reason to exist beyond co-op base building. For new players, the 1.0 release also means the game is no longer technically in development — the Early Access disclaimer is gone and the price reflects it.

On June 30, 2026, Google shipped two new generative-AI models simultaneously: Nano Banana 2 Lite, its fastest and cheapest image-generation model, and Gemini Omni Flash, a video-generation and conversational-editing model now open to developers for the first time. Together they form a text-to-video pipeline that runs entirely within the Gemini API — no third-party model hops required.

Nano Banana 2 Lite: The Price-Speed Leader

The original Nano Banana (gemini-2.5-flash-image) already undercut most alternatives on cost. Nano Banana 2 Lite cuts that further to $0.034 per 1,000-pixel image, generating results in roughly four seconds. At production scale — thousands of image calls per day — that pricing difference becomes significant. The model is available via Google AI Studio, the Gemini API, and the Gemini Enterprise Agent Platform, and is actively rolling out to consumer surfaces: AI Mode in Search, the Gemini app, NotebookLM, Google Photos, and Google Flow.

Every output automatically embeds SynthID watermarking, Google's invisible provenance signal for AI-generated imagery.

Gemini Omni Flash: Developer Video Generation, Finally

Until now, Google's Veo-based video generation was mostly locked to consumer products. Gemini Omni Flash is priced at $0.10 per second of video output, accepts text, image, and video inputs, and currently generates clips up to 10 seconds, with longer durations coming soon. Its standout feature is conversational editing: instead of regenerating from scratch when a result is close but not right, you can give follow-up instructions and the model refines the existing clip.

Google says Omni Flash matches Veo 3.1 Fast on price-performance, positioning it as the practical entry point before the heavier Veo 3.1 Standard tier.

The Combined Pipeline

Google is specifically promoting a chained workflow: generate a still with Nano Banana 2 Lite, then pass it as a reference frame to Gemini Omni Flash to animate it. The image feeds directly into the video model's context — the two integrate natively inside the Gemini API, making the end-to-end text → image → animated video flow a single-provider operation. For developers building marketing automation, social content pipelines, or interactive storytelling tools, this closes a gap that previously required stitching together separate vendors.

Both models are available today in Google AI Studio and the Gemini API. The combined release signals Google's intent to own the full generative-media stack at every price tier, from rapid prototyping to production-scale pipelines.

Vercel dropped the Next.js 16.3 Preview on June 26, 2026, and followed up with a dedicated Turbopack deep-dive on June 29. Two headline features stand out: Instant Navigations, a rethink of client-side routing that makes page transitions feel genuinely instantaneous, and a smarter Turbopack that finally stops eating your RAM on large projects.

Instant Navigations: Zero-Latency Route Transitions

The biggest user-facing improvement in 16.3 is Instant Navigations. The system uses partial prefetching to create a reusable route shell — a cached layout skeleton stored on the client. When a user clicks a link, the shell renders immediately while the full page data streams in behind it. The result is that navigating between pages feels essentially instant, even on slow connections, because the chrome of the interface is already present.

This is a meaningful evolution from earlier Next.js prefetching approaches, which pre-fetched entire pages and became expensive at scale. The new model is selective and client-cached: repeated visits to the same route get cheaper over time rather than more expensive. For content-heavy apps with large layouts, this is a genuine performance win without any code changes.

Turbopack: Memory Eviction and Persistent Cache

Turbopack has two long-requested improvements in this release. Memory eviction means the bundler now sheds build artifacts it is unlikely to need again — capping memory growth on large codebases, which was a pain point that drove some teams back to webpack. The persistent build cache survives process restarts, meaning cold starts after a deployment or CI run are dramatically faster.

Also landing: Rust React Compiler support (compiling React's new compiler in Rust for faster transforms) and import.meta.glob, which makes file-system-based dynamic imports simpler in Turbopack projects. Together these close most of the remaining parity gaps between Turbopack and webpack-based setups that still held some teams back.

AI Developer Tooling Built In

16.3 also ships a set of AI developer tools as first-class features: bundled documentation accessible inside the IDE, Skills for common Next.js patterns, browser introspection for debugging component state, and actionable error messages that suggest specific fixes rather than pointing at raw stack traces.

This reflects Vercel's bet that AI-assisted development will become the default workflow — building the scaffolding directly into the framework rather than expecting developers to assemble it from separate tools. Whether or not you use those features immediately, having them in the framework means they are ready when your team adopts them.

When to Upgrade

16.3 is in preview now, with a stable release expected within weeks. For projects already on Next.js 16.x, testing Instant Navigations and the Turbopack memory improvements in preview is worth doing early — both have a direct impact on end-user performance and daily developer experience that you will notice immediately.

id Software's follow-up content for DOOM: The Dark Ages is nearly here. Xbox Wire published a hands-on preview of the Revelations expansion on July 1, confirming it drops on July 7, 2026 — and delivering the first detailed look at its headline new weapon: the Chain Spear.

Bigger Than Both of Eternal's DLCs Combined

Revelations runs 10–12 hours of new content, making it larger than DOOM Eternal's two expansion campaigns put together. The expansion introduces new environments, enemy types, and a narrative chapter set in the aftermath of The Dark Ages' main story. This is not a map pack — it's a substantial follow-up campaign that id Software has clearly treated as a full creative undertaking.

The Chain Spear Changes How Combat Moves

The new weapon takes over the shield slot and rethinks it entirely. Rather than a defensive tool with occasional throws, the Chain Spear is built for aggressive offense and movement. Players can whip it forward to impale enemies, yank them in for follow-up melee, or use the chain itself to slingshot across gaps. According to Game Rant's preview coverage, the Chain Spear fundamentally changes the pacing of encounters — adding a grapple-like mobility option that opens traversal routes the base game's shield could never reach.

It's a smart design move. The shield was The Dark Ages' defining addition, but it locked players into a particular rhythm. The Chain Spear breaks that rhythm open while preserving the game's commitment to aggressive, movement-forward combat.

Free Update for Everyone: Ripatorium 3.0

Alongside Revelations, id Software is shipping the free Ripatorium 3.0 update on July 7 for all owners of The Dark Ages. Based on the naming pattern, this appears to be a third wave of challenge arena content building on the endgame mode introduced at launch. Precise details have not been fully disclosed yet, but the update requires no purchase — it ships to everyone on day one.

Price, Platforms, and Pass

Revelations is priced at $19.99 as a standalone purchase and is included in the Year One Pass for players who bought in at launch. It launches simultaneously on Xbox Series X|S, PC via Xbox Play Anywhere, and PlayStation 5 on July 7. If you have not started The Dark Ages yet, this week is a good time to correct that.

On June 30th, 2026, Anthropic launched Claude Sonnet 5, making it the new default model for all Free and Pro plan users on Claude.ai. It is the most capable mid-tier model the company has shipped — and arguably the most significant efficiency leap in Anthropic's lineup to date.

Near-Flagship Performance at Mid-Tier Prices

Sonnet 5's headline claim is that its performance sits close to Opus 4.8, Anthropic's top model, but at substantially lower cost. Benchmarks show marked improvements over Sonnet 4.6 on reasoning, software coding, tool use, and knowledge-intensive tasks. It's the model Anthropic is betting enterprises will reach for when they want Opus-level quality without the Opus-level invoice.

For developers, the introductory API pricing runs $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to standard pricing at $3 and $15 respectively. That positions Sonnet 5 as the obvious first choice for high-volume agentic workloads — tasks that were previously priced out of reach using Opus.

Built for Agents First

What distinguishes Sonnet 5 from its predecessor is not just raw benchmark scores — it's how the model is designed to operate autonomously. According to Anthropic, Sonnet 5 is built to be the most agentic Sonnet model yet — capable of making multi-step plans, using tools like browsers and terminals, and running long autonomous tasks at a level that previously required larger, more expensive models.

This is the real story. As AI pipelines mature, the cost of the underlying model increasingly determines whether an agentic product is viable at scale. A model that can plan, execute, and self-correct across dozens of steps, at $2 per million input tokens, changes the math for a lot of developers.

Safety by Default

Safety assessments show that Sonnet 5 has an overall lower rate of undesirable behaviors than Sonnet 4.6 and is generally safer to deploy in agentic contexts. Cybersecurity safeguards are enabled by default, with real-time detection and blocking of dangerous activity — the same protections available in Opus 4.7 and 4.8. Evaluations also show the model has significantly reduced capability to assist with offensive cybersecurity tasks compared to current Opus models.

For enterprises deploying agents that browse the web, run code, or interact with external APIs, built-in safety rails reduce the compliance friction that tends to slow adoption.

What It Means in Practice

Claude Sonnet 5 is available now across all Claude plans and via the API using the model ID claude-sonnet-5. It is also available to Max, Team, and Enterprise users alongside the existing Opus 4.8 option. If you have been holding back agentic projects because Opus pricing made them impractical at scale, the window to revisit that calculation just opened.

It has been 25 years since Master Chief crash-landed on a Forerunner ring world and changed console gaming forever. On July 28, 2026, Halo Studios is delivering a full remake of that story built from scratch in Unreal Engine 5 — and for the first time in franchise history, PlayStation 5 players get to experience it from launch day alongside Xbox and PC.

What Campaign Evolved Actually Is

Halo: Campaign Evolved is not the Anniversary remaster from 2011. It is a ground-up rebuild by Halo Studios — the development team formerly known as 343 Industries — using Unreal Engine 5's Nanite geometry and Lumen global illumination systems. All ten missions from the original Halo: Combat Evolved are present, redesigned with the updated engine, revised enemy behaviors, new weapons, and a narrative thread that connects to where the Halo universe has since expanded.

Beyond the rebuilt campaign, Campaign Evolved adds three entirely new story missions under the banner Operation: METEORITE. Set one year before the events of Combat Evolved, these missions follow Master Chief and Sergeant Avery Johnson as they infiltrate a Covenant agricultural ship in orbit above the glassed human colony of Promise, hunting for data that could lead to the Covenant's homeworld of High Charity. The prequel missions introduce new enemy types — the game refers to Sacristan warriors — and expand lore that has only been touched in peripheral novels. According to the official Halo Waypoint page, Operation: METEORITE was designed to feel native to Combat Evolved's era rather than a modern grafting-on.

Halo on PlayStation — For the First Time

The platform list is the biggest structural news. Campaign Evolved ships simultaneously on Xbox Series X/S, Windows PC via the Microsoft Store and Steam, and PlayStation 5. Full cross-play and cross-progression are supported across all three platforms, meaning a PS5 player can join an Xbox co-op session without friction, and save data transfers across platforms freely.

This is Halo on PlayStation, officially, for the first time in franchise history — a line that would have seemed impossible as recently as five years ago. It is also a signal that Xbox Game Studios has committed to multiplatform releases for its core franchises in a way that appears irreversible.

Four-Player Co-Op and Split-Screen Return

Campaign Evolved supports up to four players online in cooperative play throughout the full campaign, including the Operation: METEORITE prequel missions. Console versions also feature local two-player split-screen — a capability that was controversially absent from Halo 5 and present but limited in Halo Infinite, and one that has been actively requested by the community for years. There is no competitive multiplayer in this release. Campaign Evolved is explicitly campaign-only, a deliberate scope decision from Halo Studios.

Pricing, Game Pass, and Early Access

The Standard Edition is $49.99. A Premium Edition at $69.99 includes five days of early access starting July 23 at 8 AM PDT, plus additional cosmetic content. A Collector's Edition at $199.99 adds physical goods. All editions are available on Xbox Game Pass Ultimate and PC Game Pass from launch day, with rollout to lower-tier Standard and Premium Game Pass plans within a year.

PC system requirements are steep: Halo Studios has listed an RTX 4080 and 32GB of RAM for 4K/Ultra at 60fps — a product of UE5's rendering systems trading lower minimums for a substantially higher fidelity ceiling. On consoles, the game targets 4K/60 in quality mode and 1080p/120 in performance mode on both Xbox Series X and PS5.

The Franchise Reset Halo Needs

Halo as a franchise spent much of the 2010s trying to recover its identity after Halo 5's divisive story and Halo Infinite's uneven launch. Returning to the source — remaking the game that defined the franchise and introducing it to an entire generation that grew up on PlayStation — is a bet that the original Combat Evolved's world and characters remain compelling enough to anchor a full-price release in 2026. With Game Pass day-one inclusion, the barrier to entry is genuinely low. Whether it becomes the franchise reset Halo Studios is aiming for will be clear on July 28.

Linux 7.2-rc1 landed on June 28, 2026, with Linus Torvalds describing the merge window as "reasonably normal" — kernel mailing-list language for a clean, well-managed merge window with genuinely impactful changes. The headline addition is AMD HDMI 2.1 FRL support in the amdgpu driver, a long-overdue feature that finally brings the open-source driver in line with what RDNA 4 GPUs have been physically capable of for years.

AMD HDMI 2.1 FRL: What It Actually Changes

Until now, Linux users with AMD GPUs were effectively capped at HDMI 2.0 signaling through the open-source driver — 4K/60Hz maximum over HDMI, even on hardware designed for more. Linux 7.2-rc1 adds Fixed Rate Link (FRL) support, the HDMI 2.1 signaling technology that replaces older TMDS encoding. The practical result: AMD GPU owners on Linux can now drive uncompressed 4K at 120Hz, 8K at 60Hz, and true dynamic HDR over standard HDMI cables — no proprietary workarounds required.

This matters beyond gaming. Creators working with high-frame-rate 4K previews, anyone using a high-refresh HDR monitor over HDMI rather than DisplayPort, and TV-connected Linux setups all benefit from native support. The feature targets RDNA 4-era GPUs specifically, though related HDMI signaling improvements land across the amdgpu driver stack.

AMD ISP4: Webcam Quality Gets a Real Driver

The same release adds support for AMD's fourth-generation Image Signal Processor (ISP4), the dedicated hardware block that handles real-time camera image processing on AMD SoC platforms. On laptops powered by AMD chips, ISP4 is responsible for noise reduction, HDR stitching, and auto-focus — processing that previously required firmware workarounds or simply didn't function under Linux. Native ISP4 driver support means AMD-powered laptops should see immediate improvements in webcam quality through any standard application, without requiring vendor-specific software stacks.

Post-Quantum Signatures for Firmware Integrity

Linux 7.2-rc1 adds post-quantum ML-DSA signature support to the IMA (Integrity Measurement Architecture) and EVM (Extended Verification Module) subsystems. These components verify that boot-time firmware and loaded kernel modules haven't been tampered with. Adding ML-DSA — a lattice-based signature scheme standardized by NIST as part of its post-quantum cryptography project — means the kernel now supports signing and verification resistant to attacks from future quantum computers. For security-sensitive deployments in government, finance, and critical infrastructure, this is a practical, forward-looking addition that doesn't require waiting for a separate security patch.

strncpy Is Finally Out

One of the less-publicized but structurally significant changes is the complete removal of strncpy from the kernel codebase — the conclusion of a multi-year cleanup effort spanning over 360 individual patches. strncpy is a C function with notorious behavior: it doesn't null-terminate when the destination buffer is exactly the right size, and it zero-pads in ways that create subtle buffer overflow vulnerabilities. Eliminating it from the kernel removes an entire category of bug-prone code that has contributed to real CVEs over the years.

43 Million Lines and a Stable Release in August

The merge window pushed the Linux kernel source tree past 43 million lines — 43,179,595 by raw count, according to LinuxTeck's analysis. Linus noted the window was large but manageable, and the candidate cycle is expected to run eight weeks, with 7.2-rc2 due July 5 and a stable release projected around August 23, 2026.

That timeline puts Linux 7.2 in line to power Fedora 45 and Ubuntu 26.10 as their base kernel. For AMD GPU users in particular — especially anyone running a high-refresh monitor over HDMI or a laptop with a built-in webcam — 7.2 is the most meaningful kernel upgrade for AMD hardware in recent memory.

When reports emerged in mid-June that Qualcomm was in advanced talks to acquire Tenstorrent for up to $10 billion, the AI chip world paid close attention. The strategic logic was obvious: Qualcomm needed an AI inference engine capable of challenging Nvidia, and Tenstorrent's Blackhole chip — built on the open RISC-V instruction set — looked exactly like that. Then, on June 30, CEO Jim Keller walked into a Tokyo media event and shut it all down.

"We are not in discussions with Qualcomm about an acquisition," Keller said. He emphasized that Tenstorrent is focused on building its own business — including IP licensing and high-performance scalable AI infrastructure work across Japan and other markets. He left the door open for "partnerships in strategic areas," but the message was unambiguous: Tenstorrent is not for sale.

What Qualcomm Was Actually After

The acquisition thesis made real sense. Qualcomm was reportedly circling Tenstorrent at a $8–10B valuation — a RISC-V power play designed to crack Nvidia's grip on the AI inference market. Tenstorrent's Blackhole chip, which reached general availability on April 28, 2026, is architecturally different from Nvidia's H100s and B200s. Rather than stacking GPU parallelism, each Tensix core bundles five RISC-V processors alongside dedicated matrix and vector engines connected through a mesh-network fabric. Keller has argued this design handles the real bottleneck of modern AI inference — memory bandwidth and interconnect efficiency — better than the GPU model can.

Qualcomm, for its part, has been spending aggressively to escape its smartphone-chip reputation. On June 24, the company announced a $14 billion bet on RISC-V and an open compiler stack for AI inference — a direct attempt to undercut Nvidia's proprietary CUDA ecosystem. Tenstorrent's architecture would have been a ready-made piece of that strategy.

Building for an IPO, Not an Exit

Keller has made no secret of where he wants to take Tenstorrent. In recent interviews, he's described wanting to challenge Cerebras, Nvidia, and the broader AI hardware market on the company's own timeline. The commercial momentum backs that up: Tenstorrent recently shipped its largest order to date — a 96-Galaxy cluster comprising 3,072 Blackhole chips destined for a customer outside the US. That's not a startup scrambling for liquidity. That's a company filling serious order books.

The denial also cooled Qualcomm's deal prospects publicly, with analysts noting the deal now looks considerably less likely. For Qualcomm, that means its $14B open RISC-V infrastructure bet has to stand on its own — without Tenstorrent's chip architecture as a shortcut.

What It Means for AI Inference

Whether or not the acquisition talks were real, the story confirms that the AI inference chip market is heating up fast enough to generate $10B rumors around a company that only shipped its flagship chip three months ago. Nvidia still dominates AI training with the Blackwell platform, but inference — where the majority of enterprise AI spending is now concentrated — is increasingly contested territory. Tenstorrent, Cerebras, AMD, and a highly motivated Qualcomm are all fighting for that market.

For developers and enterprises evaluating inference infrastructure, the takeaway is that meaningful competition to CUDA is becoming real. Tenstorrent's open RISC-V approach means customers adopting Blackhole hardware aren't locked into a proprietary instruction set. That's a genuinely different value proposition from every GPU vendor in the market — and one Jim Keller clearly wants to pursue as a public company, not as a division of Qualcomm.

Open-source firmware gets a quarterly release, and the June 2026 edition of coreboot reaches notably far ahead. Coreboot 26.06, released June 25, 2026, adds early support for Intel Nova Lake and AMD Strix Halo — both silicon platforms that haven't hit consumer shelves yet — while welcoming 31 new mainboards to the project's compatibility list.

Intel Nova Lake, Early

Intel engineers contributed early Nova Lake SoC support to the coreboot tree ahead of the platform's commercial debut. It's marked as a work in progress — DDR5 isn't yet supported, and some FSP (Firmware Support Package) workarounds remain while upstream fixes are pending — but having open-source firmware support merged before the chip ships is exactly the kind of ahead-of-time engineering that makes coreboot valuable. Security-sensitive deployments won't need to wait for proprietary firmware to be auditable.

AMD Strix Halo (Ryzen AI Max 300)

AMD's Strix Halo platform — the Ryzen AI Max 300 series — also lands in 26.06. Support currently targets only the AMD Maple reference board, and like the Nova Lake work, it isn't intended for production hardware yet. But Strix Halo is AMD's next-generation integrated graphics architecture with substantial on-chip AI acceleration, and having coreboot support in the tree this early signals a productive upstream relationship between AMD and the coreboot community.

31 New Boards

The board expansion is where most users will find immediate practical value. The 31 newly supported mainboards span a wide range:

• Framework Laptop: additional Framework models join the list, extending the modular laptop's open-source firmware story to more configurations
• ASUS and ASRock: enthusiast-grade desktop boards from both manufacturers, expanding coreboot's reach into mainstream consumer hardware
• Google Chromebooks: several new Google reference devices, which use coreboot as their production firmware
• AMD Crater and AMD Jaguar: AMD reference platforms for embedded and industrial deployments

AMD ROM Armor 2

One of the more significant security improvements in 26.06 is AMD ROM Armor 2 support, which introduces A/B recovery infrastructure for failed firmware loads. When a firmware update goes wrong, the system falls back to the previous known-good image rather than bricking. On embedded and industrial hardware where hands-on recovery isn't always an option, this resilience matters significantly — and it's the kind of feature proprietary UEFI firmware has had in some implementations for years, now arriving in open-source form.

Why Coreboot Still Matters

Proprietary UEFI firmware is effectively unauditable by the people running it. It executes before the operating system, often has network access during initialization, and is implicitly trusted by everything that loads afterward. Coreboot replaces that black box with open, inspectable code that any developer or organization can read, modify, and verify.

The project's quarterly cadence with consistent Intel and AMD engineering contributions suggests coreboot is no longer a niche concern for embedded developers — it's becoming infrastructure for organizations and individuals who take firmware trust seriously. Framework Laptop support in particular means everyday Linux users can run open-source firmware on consumer hardware without buying specialist equipment. That's a meaningful shift in accessibility.

AI models can now scan an entire open source repository and surface exploitable vulnerabilities in minutes. That capability — once the domain of well-resourced security researchers — is rapidly becoming available to anyone with API access to a frontier model. The Linux Foundation's response, launched June 25, 2026, is called Akrites: a coordinated industry effort to build defenses before the offensive capability outruns the defensive one.

The Problem Akrites Is Solving

Open source software underpins nearly everything — banking systems, hospital infrastructure, power grids, telecom networks, and AI labs themselves all run on shared libraries maintained by small, often underfunded teams. Historically, the bottleneck on attacking these projects was the attacker's manual effort: reading code, understanding context, reasoning about exploitability. That bottleneck is evaporating fast.

Frontier AI models trained on codebases can identify memory corruption bugs, authentication bypasses, and injection vulnerabilities at machine speed. Once access to these capabilities is broadly available, bad actors who previously lacked the technical expertise to mount sophisticated attacks will have the tools to do it automatically. The window for proactive defense is now.

What Akrites Actually Does

The Linux Foundation's announcement establishes two concrete structures: a shared Security Incident Response Team (SIRT) staffed by participating organizations, and a standardized Coordinated Vulnerability Disclosure (CVD) process built on confidentiality-first principles. The idea is to create a single trusted pipeline for reporting, reproducing, remediating, and disclosing vulnerabilities in critical open source projects — with enough staffing and tooling behind it to actually act on findings quickly rather than letting them sit in maintainer inboxes for months.

Seed funding comes from Alpha-Omega, a directed fund of the Linux Foundation that already supports security work across the open source ecosystem.

Who's In

The founding member list is notable for its breadth. Amazon Web Services, Anthropic, Chainguard, Cisco, Citi, Endor Labs, Ericsson, Google, IBM, JPMorganChase, Microsoft and GitHub, NVIDIA, OpenAI, RapidFort, Red Hat, the Rust Foundation, Sonatype, Vodafone, and Zscaler all signed on at launch — 19 organizations spanning cloud providers, AI labs, financial institutions, telecoms, and open source stewards, all agreeing that the threat is real enough to set competition aside and collaborate.

The AI labs' participation is worth noting specifically. Anthropic and OpenAI are among the organizations whose models could theoretically be weaponized for the kind of AI-assisted vulnerability scanning Akrites is designed to defend against. Their presence as founding members represents an acknowledgment that they bear some responsibility for funding the defensive infrastructure their tools could stress-test.

The Timing

Akrites arrives in the same month that AI security research teams demonstrated finding tens of thousands of real vulnerabilities in critical infrastructure codebases using AI-assisted scanning techniques. The capability isn't theoretical — it exists and is being used. The open source community's response to date has been fragmented and chronically underfunded. Akrites is an attempt to change that structural problem with coordinated institutional support.

Whether the project develops real staying power will depend on how much engineering time the founding members actually dedicate versus how much is press-release participation. The structure looks right. The test is follow-through.

Valve's Steam Machine has gone from regulatory filing to living-room console in the span of a few months — and on June 29, 2026, it officially ships. The launch details are now confirmed: a compact cube-shaped Linux PC starting at $1,049, designed to bring the entire Steam library to your TV without the compromises of a traditional locked-down console.

What's Inside

Under the lid is a six-core, 12-thread AMD Zen 4 processor boosting up to 4.4 GHz, paired with a semi-custom RDNA 3 GPU carrying 28 compute units and 8 GB of dedicated GDDR6 video memory. Shared system memory comes in at 16 GB DDR5. Storage options are 512 GB and 2 TB, with a TDP target that keeps the box quiet enough for a living room setup.

Valve claims the Steam Machine delivers roughly six times the GPU performance of the original Steam Deck — a gap wide enough to run modern titles at 4K with high-fidelity settings rather than the handheld's compromised medium-ish presets.

A Console That Isn't

The design philosophy is the unusual part. Rather than a closed platform, the Steam Machine runs SteamOS — Valve's Linux-based operating system — but ships as an open platform. Users can install other storefronts, switch to desktop mode, sideload applications, or run non-Steam software. The box boots directly into the Steam interface, handling cloud saves and fast suspend-and-resume the way a console would, but without the walled garden.

That means the Heroic Games Launcher for Epic and GOG titles, Lutris for legacy libraries, and RetroArch for emulation all work out of the box. The Steam Machine is, functionally, a small-form-factor PC wearing a console's costume — and Valve is betting that's exactly what a meaningful slice of the market wants.

Pricing and How to Buy

Valve used a reservation-based system for the launch wave. Signups closed June 25 at 10 a.m. PT, and Valve randomized the list before sending purchase invitations. First units ship June 30.

• 512 GB, no controller: $1,049
• 512 GB with Steam Controller: $1,128
• 2 TB, no controller: $1,349
• 2 TB with Steam Controller: $1,428

The pricing is firmly PC-tier rather than console-tier. Sony and Microsoft sell their current consoles at $499–$599 and subsidize hardware with software royalties. Valve doesn't take a cut of third-party storefronts running on its hardware, so the margin math is different — you're paying a fair price for a small PC, not a loss-leader device.

Why This Matters

FCC filings from earlier this year hinted at the June timeline, but the actual reservation process revealed real demand: hundreds of thousands of signups for a $1,000+ gaming appliance running Linux.

The Steam Deck normalized handheld Linux gaming for mainstream players. The Steam Machine is the next step: a couch-friendly, TV-connected device running the same OS but with proper desktop-class cooling, full keyboard-and-mouse support when you want it, and none of the portability-driven compromises that made the Deck's performance a constant negotiation with game settings.

Valve has spent years building Proton compatibility to the point where the overwhelming majority of the Steam catalog runs without manual configuration. That investment is the foundation the Steam Machine is built on — and today it ships.

When Meta launched its first AI-powered smart glasses in 2023, they came in partnership with EssilorLuxottica and wore the Ray-Ban logo. Three years and two generations later, Meta has its own frames. On June 23, 2026, Meta introduced Meta Glasses — three new styles starting at $299, no partner branding required.

The lineup

The launch includes three models: Adventurer, Fury, and a limited-edition Meta Glasses by Kylie — designed in collaboration with Kylie Jenner in her first wearable tech venture. All three include a built-in camera, open-ear speakers, and Meta AI powered by Muse Spark, Meta's internal multimodal model. At $299, the entry price is at least $80 lower than Meta's second-generation Ray-Ban glasses.

The glasses went on sale June 23 in the United States, Canada, the United Kingdom, France, Italy, Germany, Spain, and several other European markets.

Meta AI on from day one

What Meta is emphasizing most is that these glasses ship with Meta AI fully active at launch — not added later via a software update. The glasses can see what you see, answer questions about your environment, identify objects, translate text, and handle hands-free calling with real-time AI assistance. Meta AI runs on a combination of on-device processing and cloud inference through Muse Spark.

The capability set is a step forward from the original Ray-Ban Meta glasses, which launched with much more limited AI functionality and added features gradually over time. Shipping a complete AI experience from day one reflects how much the underlying model infrastructure has matured in two years.

The market context

According to Counterpoint Research, Meta currently holds approximately 80% of the AI smart glasses market — a category that barely existed two years ago. Snap's competing Specs, announced at Augmented World Expo earlier in June, come in at $2,195 — more than seven times the price of Meta's entry model. Meta is betting that driving down cost is the fastest way to lock in the category before competitors can.

EssilorLuxottica remains a manufacturing partner for Meta Glasses, even as Meta introduces its own brand identity. The two companies also separately announced a lens development collaboration with Applied Materials, suggesting the hardware partnership isn't going anywhere — only the logo has changed.

What this signals

Meta launching under its own brand name is a quiet but significant move. It puts the product identity directly on Meta rather than licensing Luxottica's brand equity. If AI glasses become as mainstream as Meta is betting, Meta wants its name on the category — not a fashion brand it doesn't fully control. The $299 price point is the first real test of whether consumers will buy AI glasses at mass-market prices, and whether a Meta logo carries enough appeal to sell them without a Ray-Ban co-sign.

Open source software powers much of the world's critical infrastructure — financial systems, hospitals, government services — and most of it is maintained by volunteers who patch vulnerabilities when they can. IBM and Red Hat launched Project Lightwell in May 2026 with a $5 billion commitment and 20,000 dedicated engineers to change that picture. On June 26, 2026, they announced that Deloitte is joining as an integration collaborator.

What Project Lightwell actually does

Lightwell operates differently from a traditional vulnerability scanner. Rather than simply flagging CVEs and leaving remediation to you, it coordinates upstream disclosures with independent open source maintainers, writes and validates backported patches, and delivers tested fixes directly to the specific, pinned software versions running in customers' production environments — without requiring a full library upgrade.

That last part matters enormously in enterprise environments where upgrading a dependency version is often weeks of testing and approvals. Lightwell sidesteps the problem by meeting software where it already is.

The system works in three stages:

• Continuous visibility: Mapping and scanning first-party, open source, and third-party components to track exactly what runs where.
• Contextual prioritization: Scoring active threats against severity, exploitability, and threat-chaining to separate signal from noise.
• Machine-speed remediation: Automated patch validation from Red Hat and IBM, deployed into production repositories with minimal disruption.

Deloitte's role

Deloitte joins as an integration collaborator, bringing its secured software supply chain architecture and cyber risk services. Practically, Deloitte will maintain a bench of Forward Deployed Engineers who work directly inside client environments to support ongoing remediation. IBM and Red Hat handle the engineering at scale; Deloitte handles the organizational integration and coordination.

Palo Alto Networks also joined on June 24, adding network-level threat protection that combines with Lightwell's software remediation. An emerging threat gets blocked at the perimeter while a validated patch is being prepared — two layers of protection running in parallel.

Who's already in

The project counts Bank of America, Citi, Goldman Sachs, JPMorganChase, Mastercard, Morgan Stanley, Royal Bank of Canada, State Street, Visa, and Wells Fargo among its early adopters — a signal that financial services is the immediate target market, where both regulatory obligations and attack-surface exposure are exceptionally high.

With Deloitte and Palo Alto now added to IBM and Red Hat's coalition, Lightwell is starting to look less like a product and more like a new kind of infrastructure for enterprise open source. Whether it can actually keep pace with the attack surface — which grows faster than any single team can patch — is the real question.

OpenAI shipped a meaningful expansion to its Codex product on June 27, 2026: Codex Remote is now generally available on all ChatGPT plans. Previously limited to a closed preview, the feature lets any subscriber use the ChatGPT mobile app to connect to a Mac or Windows machine, kick off or resume coding tasks, and approve actions from their phone.

The experience is built around OpenAI's agentic Codex model, which can read files, run code, write tests, and navigate terminals — the same capabilities available in the desktop app. The difference is that you no longer need to be sitting at your computer. You can start a task on your lunch break or approve a pull request from the couch, with the heavy lifting happening on your development machine back at the desk.

How the pairing works

Remote Control uses authenticated QR pairing between your iOS or Android device and your Mac or Windows host. Each pair is one-to-one — a phone connects to a specific machine, not a shared endpoint. Connections set up since June 8, 2026 carry over automatically; older inactive connections need to pair again.

OpenAI also added a DigitalOcean Droplet plugin at launch. If you'd rather not run Codex against your local machine, you can provision a fresh cloud server, configure SSH access, and connect it to the Codex app as a remote workspace — all from inside ChatGPT.

Why this matters

For developers who already use Codex for longer-running tasks — full feature implementation, test suite generation, codebase refactors — Codex Remote removes a key friction point. Instead of staying anchored to a desk to monitor progress, you can check in from anywhere and approve the next step without losing momentum.

The ChatGPT Business plan also got a simplified model picker alongside this release, making it easier to switch between speed-optimized and reasoning-heavy models. Separately, GPT-4.5 retired from ChatGPT on June 26, completing a migration that began with the May 28 sunset notice — existing conversations automatically moved to the corresponding GPT-5.5 model.

Codex Remote is available today for all ChatGPT plan tiers on both Mac and Windows. Full documentation and the DigitalOcean plugin setup are in the OpenAI Codex developer docs.