A claim is spreading fast in hardware circles: that Google’s newest coding-focused AI model, Gemini 4 Argon, was used to convert Alan Wake 2, a graphically demanding Windows-only title, into a native Arm64 app running on Apple’s M3 Ultra chip at over 140 frames per second, with no Rosetta translation layer involved. The claim traces back to a single source: developer Max Weinbach, who posted the experiment on X. Pakistani outlet ProPakistani picked it up on October 15, 2026, framing it as Gemini 4 Argon turning Alan Wake 2 into “a native MacBook game.”
That framing has outpaced the facts on the ground. No independent reviewer has reproduced the port. No benchmark lab has verified the 140 fps figure, the test resolution, the graphics settings, or whether frame times stayed stable rather than spiking. What exists, as of October 16, 2026, is one developer’s account of a side project built on top of a Google AI model that Google itself markets as a coding and problem-solving tool, not a game-porting product. This piece lays out exactly what is known, what is claimed, and what it would mean for Apple Silicon, Windows gaming, and the AI coding-agent race if the claim holds up.
What Max Weinbach Actually Claimed
According to the post Weinbach shared, and as relayed by ProPakistani, the experiment involved taking Alan Wake 2, Remedy Entertainment’s survival-horror title built for x86 Windows machines, and converting it to run natively on Arm64 macOS. Weinbach described the project in his own words: “My favorite Gemini 4 Argon experiment was converting Alan Wake 2 from x86 and Windows to native Arm64 on macOS,” a quote reported by ProPakistani.
The technical steps he described break into four parts. First, writing Rust bindings to stand in for x86-specific instructions and operations that Arm64 silicon does not execute natively. Second, converting the Windows-side game logic to something macOS can run. Third, rewriting the game’s graphics shaders from whatever Windows API Alan Wake 2 targets into Apple’s Metal shading language. Fourth, compiling the whole thing ahead of time, skipping Rosetta 2, Apple’s standard x86-to-Arm translation layer for running Intel-built apps on Apple Silicon Macs.
Each of those four steps, done properly, is a substantial engineering project on its own. Shader translation alone has tripped up professional porting studios for years, which is part of why Nvidia’s own DLSS 5 workaround for Apple Silicon only managed single-digit frame rates when it surfaced in September. Doing all four, through an AI coding assistant, on what reads like a personal side project, and landing at a reported 140 fps on the hardware Weinbach used, an M3 Ultra, is the part straining credulity among engineers who’ve seen this kind of claim before.
What Gemini 4 Argon Actually Is
Separate from the Alan Wake 2 story, Gemini 4 Argon is a real, officially shipped Google product. Google’s own blog post, published on its Indonesian regional site and describing the model’s design goals, pitches it as built to tackle the most complex class of problems the company’s AI stack has attempted, and the model is available now to subscribers on Google’s AI Pro and AI Ultra consumer tiers.
Google has also pushed Gemini 4 Argon into developer tooling well beyond the Gemini app. The company’s Android developer blog detailed bringing the model into Android’s agentic engineering workflow and its Antigravity CLI, positioning Argon less as a chatbot and more as a tool meant to sit inside a developer’s terminal, refactor code, and act with some autonomy across a codebase. That framing matters here: if Argon is genuinely capable of multi-step autonomous code transformation, a determined developer attempting an x86-to-Arm64 game port isn’t an absurd use case. It’s a stress test of exactly the capability Google is marketing.
Hong Kong-based outlet Futu News covered the model’s broader rollout too, reporting on Google opening Gemini 4 Argon globally for code refactoring tasks. Google’s DeepMind unit, which develops the Gemini family, maintains its own model index at deepmind.google, where Argon sits alongside the rest of the Gemini lineup rather than being marketed as a separate, gaming-adjacent product.
Shattered.io has tracked Argon’s rollout closely since late September. Our earlier coverage found the model winning only one of four head-to-head AI benchmark tests against rival models, and separately reported Argon fabricating emails during a simulated business task on the Vending Bench evaluation, ranking third rather than first. Neither data point disqualifies Argon from doing useful code conversion work. Both are reasons to treat an unverified, single-source claim about a 140 fps native game port with more scrutiny than enthusiasm.
What’s Confirmed, What’s a Claim, and What’s Missing
Separating the verified record from the viral headline matters more than usual here, because the headline racing around tech aggregators, “Gemini 4 Argon runs Alan Wake 2 native at 140 fps,” reads like an independently confirmed benchmark result. It isn’t. It’s a paraphrase of one person’s description of their own project.
| Detail | Status | Source |
|---|---|---|
| Weinbach used Gemini 4 Argon on an x86-to-Arm64 Alan Wake 2 conversion | Claimed by Weinbach | Posted on X, relayed by ProPakistani |
| Test hardware was an M3 Ultra | Claimed by Weinbach | ProPakistani |
| Performance exceeded 140 fps | Claimed by Weinbach | ProPakistani |
| No Rosetta 2 translation was used | Claimed by Weinbach | ProPakistani |
| Resolution, graphics settings, frame-time stability | Not disclosed | None |
| Independent reproduction by a reviewer or outlet | Not yet available | None |
| Gemini 4 Argon is a shipped Google product on AI Pro/Ultra | Confirmed | Google’s official blog |
| Argon is integrated into Android’s Antigravity CLI | Confirmed | Android Developers Blog |
That gap, between a confirmed product and an unverified anecdote about what someone did with it, is the entire story. It’s also a familiar pattern. Claims about AI models doing something dramatically beyond their stated purpose tend to spread on social media faster than anyone can check them, and corrections rarely travel as far as the original post.
Why a Native Arm64 Game Port Is Hard, Even With AI Help
Porting a AAA Windows game to Apple Silicon without Rosetta touches nearly every layer of the software stack. Alan Wake 2 relies on a DirectX-based rendering pipeline, Windows-specific audio and input APIs, and almost certainly some inline x86 assembly or vectorized instructions in its performance-critical paths. Converting all of that to Metal, macOS system calls, and Arm64 instructions isn’t a find-and-replace job. It’s closer to a full re-architecture of the game’s lower layers while trying to keep the gameplay, physics, and visual output identical.
That’s exactly why professional porting houses charge real money and take real time to do this work for shipped titles, and why even industry giants have struggled with partial versions of the problem. Apple’s own M5 Max Mac Studio, reviewed by shattered.io, hit 60 fps in native 4K gaming on titles built for Apple Silicon from the ground up, not converted after the fact from a Windows codebase. Nvidia’s attempt to bring its own DLSS 5 upscaling technology to Apple Silicon via a Metal workaround topped out at roughly 2 fps, a result shattered.io covered in September, underscoring how much performance typically evaporates when a Windows-native graphics pipeline gets forced onto Apple’s stack.
Against that backdrop, a reported 140 fps result from a reportedly AI-assisted conversion is either a genuinely significant engineering result or a number that reflects a stripped-down build, a low resolution, reduced settings, or a benchmark run that doesn’t resemble how most players would actually experience the game. Without disclosed methodology, there is no way to know which.
Market Impact: Apple Silicon’s Gaming Narrative Gets a Jolt
Regardless of whether the 140 fps figure holds up to scrutiny, the claim is landing at a moment when Apple’s gaming credibility on its own silicon is a live industry talking point. Apple’s M5 Pro and M5 Max chips, which shattered.io profiled after seven months on the market, pushed memory bandwidth to 614GB/s, numbers that matter for the kind of texture and asset streaming heavy game engines demand. Earlier in the M5 generation, the M5 Ultra Mac Studio posted a 90% performance jump in shattered.io’s own review, at a $12,299 price point that puts it well outside mainstream gaming budgets.
Nvidia, meanwhile, is treating Apple’s unified-memory approach as a competitive threat worth responding to directly. The company’s RTX Spark line, which shattered.io reported shipping October 16, 2026, the same week as this Alan Wake 2 claim, was built explicitly with Apple’s 128GB unified memory ceiling in mind, a sign Nvidia sees Apple Silicon’s memory capacity, not just its raw GPU throughput, as the thing it needs to match. If AI-assisted porting tools genuinely start closing the gap between Windows’s massive game library and Apple’s walled-off hardware, that changes the calculus for every GPU maker chasing the same AI-PC and creative-workstation buyers Apple has been courting.
There’s a second-order effect worth tracking too: investor and developer attention. A viral, unverified claim about AI-powered game porting is the kind of story that can move sentiment around both AI coding tools and Apple’s hardware roadmap well before anyone confirms the underlying number. Google has not commented on the Alan Wake 2 experiment specifically, and nothing in Google’s official Gemini 4 Argon material frames the model as a game-porting tool. The company’s own positioning is about general coding and problem-solving capability, with game conversion, if it happened at all, sitting as an unsanctioned side use rather than a marketed feature.
Competitive Comparison: Where Argon Sits Among AI Coding Agents
Gemini 4 Argon doesn’t operate in a vacuum. It’s one entrant in a crowded field of AI coding agents, several of which shattered.io has benchmarked against each other in recent weeks. Argon’s own track record is mixed: it won one of four head-to-head AI benchmark tests in shattered.io’s early-October coverage, and it was separately caught fabricating email outputs during the Vending Bench business-simulation test, a flaw that knocked it down to third place in that specific evaluation rather than first.
| Signal | What It Shows | Date Reported |
|---|---|---|
| Argon wins 1 of 4 head-to-head AI benchmarks | Mixed general-purpose performance versus rivals | Oct. 1, 2026 |
| Argon gated to 650 partners at launch | Controlled, staged rollout rather than open release | Oct. 1, 2026 |
| Argon pricing set at $2/$10 tiers | Positioned as a premium coding/reasoning model | Oct. 1, 2026 |
| Argon fabricates emails on Vending Bench, ranks #3 | Reliability gap on autonomous business tasks | Oct. 4, 2026 |
| Google cuts free Gemini tier days after Argon’s debut | Argon treated as a paid-tier differentiator | Oct. 4, 2026 |
| Weinbach’s Alan Wake 2 porting claim surfaces | Unverified use-case claim, single source | Oct. 15, 2026 |
That timeline matters for context. Google rolled Argon out in a gated fashion, limited to roughly 650 partners initially, then narrowed free-tier access to a single lighter model days later, a sequence that reads like a company steering usage toward paying customers rather than throwing Argon open for mass experimentation. A hobbyist game-porting project, if accurate, would be exactly the kind of unsanctioned, resourceful use case that slips through a gated rollout rather than one Google planned for or is promoting.
Historical Context: AI-Assisted Porting Has a Spotty Track Record
This isn’t the first time an AI tool has been credited, rightly or wrongly, with cracking a cross-platform porting problem that stumped larger engineering teams. Earlier in the Apple Silicon era, independent developer antirez built tooling to run large language models directly on Apple’s M5 Max chip at meaningful speeds, a project shattered.io covered as a sign that clever individual engineering, sometimes AI-assisted, sometimes not, can occasionally outpace what big vendors ship officially. Nvidia’s own Metal-based DLSS 5 workaround for Apple Silicon, despite coming from a company with vastly more engineering resources than a solo developer, landed at single-digit frame rates rather than anything playable.
The pattern across these stories is consistent: official, well-resourced porting efforts tend to produce modest, carefully measured results, while informal, individually driven projects produce the most dramatic headline numbers, and the most dramatic numbers are also the ones that go the longest without independent verification. Alan Wake 2 on Gemini 4 Argon fits that pattern precisely. It would not be the first AI-era claim to either hold up under scrutiny and reshape expectations, or quietly fail to reproduce once someone else tries it on their own M3 Ultra.
The Skepticism Developers Are Already Raising
Within hours of the claim circulating, the core objection among engineers wasn’t whether Gemini 4 Argon can refactor code, it clearly can do that at some level given Google’s own Android Antigravity CLI integration, but whether a model built for coding assistance can autonomously produce correct, performant Metal shader translations and low-level Arm64 bindings for a commercial game engine it was never trained on in that specific context. Shader compilation errors, silent precision mismatches, and subtle physics drift are the kinds of bugs that don’t show up in a quick frame-rate counter screenshot but absolutely show up during an actual playthrough.
None of that proves the claim false. It does explain why outlets outside ProPakistani’s initial report have been cautious about repeating the 140 fps figure as settled fact, and why this piece treats it as a claim under a named person’s byline rather than a confirmed benchmark.
Could This Work for Other Windows Games?
If Weinbach’s Alan Wake 2 result turns out to be reproducible and not an outlier tied to that specific game’s engine, the obvious next question is whether the same Gemini 4 Argon workflow generalizes. Alan Wake 2 runs on Remedy’s Northlight engine, a proprietary, heavily optimized codebase that behaves differently from, say, an Unreal Engine or Unity title. A conversion pipeline tuned to Northlight’s quirks, even an AI-assisted one, may not transfer cleanly to a different engine’s shader architecture or memory layout.
That’s a meaningful caveat for anyone hoping this points toward a general-purpose tool for bringing Windows-only libraries to Apple Silicon. A single successful conversion, even a verified one, demonstrates a technique works for one engine under one developer’s conditions. It does not demonstrate a repeatable product. The gap between “someone did this once” and “this reliably works across the back catalog” is exactly where past AI-assisted engineering claims have tended to stall once more people tried to replicate them.
The Business Case: Why Would Google Even Want This?
Google’s public materials give no indication the company is pursuing game porting as a Gemini 4 Argon use case, and there’s a straightforward commercial reason for that. Google doesn’t sell Mac hardware, doesn’t operate a PC game storefront, and has no obvious revenue line that benefits from more Windows games running on Apple Silicon. If anything, a tool that helps developers escape platform lock-in cuts against the interests of platform holders generally, Apple included, since Apple’s own App Store and Mac App Store economics depend on controlling what runs natively on its hardware.
What Google clearly does want is for developers to see Gemini 4 Argon as capable of hard, multi-step coding problems, which is precisely why the model’s Android Antigravity CLI integration emphasizes autonomous, agentic code work rather than narrow chatbot assistance. A viral anecdote about Argon tackling an ambitious porting project, true or not, functions as free marketing for that general capability story, even without Google lifting a finger to promote it directly.
What to Watch Next
- Whether Weinbach, or anyone else, publishes the converted build, source diffs, or a reproducible benchmark methodology rather than a single performance figure.
- Whether Remedy Entertainment or any official Alan Wake 2 rights holder comments on an unauthorized native macOS conversion of their game.
- Whether Google acknowledges or distances itself from game-porting as a use case for Gemini 4 Argon, given its current positioning around coding and agentic engineering tasks.
- Whether other developers attempt similar x86-to-Arm64 conversions on different titles using Argon or rival coding agents, and whether results cluster near 140 fps or fall well short.
- Whether Apple’s hardware roadmap, including chips beyond the current M5 generation, starts getting marketed with AI-assisted game compatibility as an explicit selling point.
Predictions: Where This Story Goes From Here
First, expect at least one independent attempt at reproduction within the next few weeks, given how much attention the claim has already drawn; whether it succeeds will determine if this becomes a genuine inflection point or a footnote. Second, Google is unlikely to formally endorse game porting as a Gemini 4 Argon feature in the near term, since its current messaging is built around coding and problem-solving rather than entertainment software compatibility. Third, Nvidia and AMD will keep treating Apple’s unified memory architecture, not AI-assisted porting specifically, as the more urgent competitive threat, since RTX Spark’s design choices already reflect that priority. Fourth, if the 140 fps figure cannot be reproduced under disclosed conditions, expect the story to fade from AI-model coverage within a month, much as several previous viral AI-capability claims have. Fifth, regardless of how the Alan Wake 2 claim specifically resolves, expect more developers to publicly experiment with AI coding agents on cross-platform porting tasks, simply because the tools are now accessible enough that trying costs little beyond time.
Frequently Asked Questions
Did Google officially port Alan Wake 2 to Apple Silicon?
No. The claim comes from developer Max Weinbach, who described using Gemini 4 Argon for a personal porting experiment. Google has not released an official Alan Wake 2 port and has not commented on the project as a sanctioned use case for the model.
Is the 140 fps figure verified by any independent source?
No. As of October 16, 2026, the figure comes solely from Weinbach’s own account, relayed by ProPakistani. No benchmarking outlet, reviewer, or lab has independently reproduced or measured the result, and resolution and graphics settings used for the test have not been disclosed.
What is Gemini 4 Argon, officially?
Gemini 4 Argon is a Google AI model available to Google AI Pro and AI Ultra subscribers, and it is also integrated into Android’s agentic developer tooling, including the Antigravity CLI. Google positions it as built for complex problem-solving and coding tasks, not as a dedicated game-porting or gaming product.
What hardware was reportedly used for the test?
Weinbach’s account names an M3 Ultra system as the hardware used. No further configuration details, such as RAM amount or specific macOS version, have been disclosed in the public reporting.
Why is converting a Windows game to native Arm64 macOS considered difficult?
It requires rewriting graphics shaders for a different API (Metal instead of the Windows-side rendering pipeline), replacing x86-specific instructions with Arm64 equivalents, and handling the conversion without Rosetta 2, Apple’s standard translation layer. Each piece is a substantial engineering task on its own, which is why full native ports of existing Windows titles remain rare outside official studio efforts.
How does this compare to Nvidia’s attempt to bring DLSS 5 to Apple Silicon?
Nvidia’s Metal-based workaround for DLSS 5 on Apple Silicon, reported by shattered.io in September 2026, achieved roughly 2 fps, far short of anything playable. That gap illustrates how significant a verified 140 fps result would be if it holds up, and why the current, unverified claim is drawing both excitement and skepticism.
Has Remedy Entertainment commented on the claim?
No public comment from Remedy Entertainment or Alan Wake 2’s publisher has been reported as of October 16, 2026.
Where can I read the original report?
ProPakistani’s October 15, 2026 article, which first relayed Weinbach’s claim in detail, is the primary public record of the experiment. Google’s official Gemini 4 Argon material and DeepMind’s model page provide the confirmed, non-gaming context for the underlying AI model.