Google I/O? More like Google AI Gemini 2.5 is leaving preview just in time for Google’s new $250 AI subscription Gemini 2.5 is rolling out everywhere, and you can pay Google $250 per month for more of it. Ryan Whitwam – May 20, 2025 5:03 pm | 44 All the new Gemini AI at I/O. Credit: Ryan Whitwam All the new Gemini AI at I/O. Credit: Ryan Whitwam Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more MOUNTAIN VIEW, Calif.—Google rolled out early versions of Gemini 2.5 earlier this year. Marking a significant improvement over the 2.0 branch. For the first time, Google's chatbot felt competitive with the likes of ChatGPT, but it's been "experimental" and later "preview" since then. At I/O 2025, Google announced general availability for Gemini 2.5, and these models will soon be integrated with Chrome. There's also a fancy new subscription plan to get the most from Google's AI. You probably won't like the pricing, though. Gemini 2.5 goes gold Even though Gemini 2.5 was revealed a few months ago, the older 2.0 Flash has been the default model all this time. Now that 2.5 is finally ready, the 2.5 Flash model will be swapped in as the new default. This model has built-in simulated reasoning, so its outputs are much more reliable than 2.0 Flash. Google says the release version of 2.5 Flash is better at reasoning, coding, and multimodality, but it uses 20–30 percent fewer tokens than the preview version. This edition is now live in Vertex AI, AI Studio, and the Gemini app. It will be made the default model in early June. Likewise, the Pro model is shedding its preview title, and it's getting some new goodies to celebrate. Recent updates have solidified the model's lead on the LM Arena leaderboard, which still means something to Google despite the recent drama—yes, AI benchmarking drama is a thing now. It's also getting a capability called Deep Think, which lets the model consider multiple hypotheses for every query. This apparently makes it incredibly good at math and coding. Google plans to do a little more testing on this feature before making it widely available. Deep Think is more capable of complex math and coding. Credit: Ryan Whitwam Both 2.5 models have adjustable thinking budgets when used in Vertex AI and via the API, and now the models will also include summaries of the "thinking" process for each output. This makes a little progress toward making generative AI less overwhelmingly expensive to run. Gemini 2.5 Pro will also appear in some of Google's dev products, including Gemini Code Assist. Gemini Live, previously known as Project Astra, started to appear on mobile devices over the last few months. Initially, you needed to have a Gemini subscription or a Pixel phone to access Gemini Live, but now it's coming to all Android and iOS devices immediately. Google demoed a future "agentic" capability in the Gemini app that can actually control your phone, search the web for files, open apps, and make calls. It's perhaps a little aspirational, just like the Astra demo from last year. The version of Gemini Live we got wasn't as good, but as a glimpse of the future, it was impressive. There are also some developments in Chrome, and you guessed it, it's getting Gemini. It's not dissimilar from what you get in Edge with Copilot. There's a little Gemini icon in the corner of the browser, which you can click to access Google's chatbot. You can ask it about the pages you're browsing, have it summarize those pages, and ask follow-up questions. Google AI Ultra is ultra-expensive Since launching Gemini, Google has only had a single $20 monthly plan for AI features. That plan granted you access to the Pro models and early versions of Google's upcoming AI. At I/O, Google is catching up to AI firms like OpenAI, which have offered sky-high AI plans. Google's new Google AI Ultra plan will cost $250 per month, more than the $200 plan for ChatGPT Pro. So what does your $250 get you every month? You'll get all the models included with the basic plan with much higher usage limits. If you're using video and image generation, for instance, you won't bump against any limits. Plus, Ultra comes with the newest and most expensive models. For example, Ultra subs will get immediate access to Gemini in Chrome, as well as a new agentic model capable of computer use in the Gemini API (Project Mariner). Gemini Ultra has everything from Pro, plus higher limits and instant access to new tools. Credit: Ryan Whitwam That's probably still not worth it for most Gemini users, but Google is offering a deal right now. Ultra subscribers will get a 50 percent discount for the first three months, but $125 is still a tough sell for AI. It's available in the US today and will come to other regions soon. A faster future? Google previewed what could be an important advancement in generative AI for the future. Most of the text and code-based outputs you've seen are generated from beginning to end, token by token. Its large language model (LLM) Diffusion works a bit differently for image generation, but Google is now experimenting with Gemini Diffusion. Diffusion models create images by starting with random noise and then denoise it to create what you asked for. Gemini Diffusion works similarly, generating entire blocks of tokens at the same time. The model can therefore work much faster, and it can check its work as it goes to make the final output more accurate than comparable LLMs. Google says Gemini Diffusion is 2.5 times faster than Gemini 2.5 Flash Lite, which is its fastest standard model, while also producing much better results. Google claims Gemini Diffusion is capable of previously unheard-of accuracy in complex math and coding. However, it's not being released right away like many of the other I/O Gemini features. Google DeepMind is accepting applications to test it, but it may be a while before the model exits the experimental stage. Even though I/O was wall-to-wall Gemini, Google still has much, much more AI in store. Ryan Whitwam Senior Technology Reporter Ryan Whitwam Senior Technology Reporter Ryan Whitwam is a senior technology reporter at Ars Technica, covering the ways Google, AI, and mobile technology continue to change the world. Over his 20-year career, he's written for Android Police, ExtremeTech, Wirecutter, NY Times, and more. He has reviewed more phones than most people will ever own. You can follow him on Bluesky, where you will see photos of his dozens of mechanical keyboards. 44 Comments

Aurora Innovation fait marche arrière. La société de conduite autonome, qui a lancé fin avril un service commercial avec des camions équipés de...

AI apps are coming Google to give app devs access to Gemini Nano for on-device AI New APIs for Google's ML Kit will let developers plug into the on-device AI model. Ryan Whitwam – May 16, 2025 2:15 pm | 16 Credit: Thomas Fuller/SOPA Images/LightRocket via Getty Images Credit: Thomas Fuller/SOPA Images/LightRocket via Getty Images Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more The rapid expansion of generative AI has changed the way Google and other tech giants design products, but most of the AI features you've used are running on remote servers with a ton of processing power. Your phone has a lot less power, but Google appears poised to give developers some important new mobile AI tools. At I/O next week, Google will likely announce a new set of APIs to let developers leverage the capabilities of Gemini Nano for on-device AI. Google has quietly published documentation on big new AI features for developers. According to Android Authority, an update to the ML Kit SDK will add API support for on-device generative AI features via Gemini Nano. It's built on AI Core, similar to the experimental Edge AI SDK, but it plugs into an existing model with a set of predefined features that should be easy for developers to implement. Google says ML Kit’s GenAI APIs will enable apps to do summarization, proofreading, rewriting, and image description without sending data to the cloud. However, Gemini Nano doesn't have as much power as the cloud-based version, so expect some limitations. For example, Google notes that summaries can only have a maximum of three bullet points, and image descriptions will only be available in English. The quality of outputs could also vary based on the version of Gemini Nano on a phone. The standard version (Gemini Nano XS) is about 100MB in size, but Gemini Nano XXS as seen on the Pixel 9a is a quarter of the size. It's text-only and has a much smaller context window. Not all versions of Gemini Nano are created equal. Credit: Ryan Whitwam Not all versions of Gemini Nano are created equal. Credit: Ryan Whitwam This move is good for Android in general because ML Kit works on devices outside Google's Pixel line. While Pixel devices use Gemini Nano extensively, several other phones are already designed to run this model, including the OnePlus 13, Samsung Galaxy S25, and Xiaomi 15. As more phones add support for Google's AI model, developers will be able to target those devices with generative AI features. The documentation is available for developers to peruse now, but we expect Google to fling the API doors open at I/O. The company has already confirmed an I/O session called "Gemini Nano on Android: Building with on-device gen AI." The description promises new APIs to "summarize, proofread, and rewrite text, as well as to generate image descriptions," which sounds exactly like what the new ML Kit APIs can do. An important piece of the AI puzzle App developers interested in adding on-device generative AI features on Android are currently in a tough spot. Google offers the AI Edge SDK that can provide access to the NPU hardware for running models, but these tools are experimental and only work on the Pixel 9 series currently. It's also limited to text. Both Qualcomm and MediaTek offer APIs for running AI workloads, but features and functionality vary by device, which makes it risky to rely on them for a long-term project. And running your own model requires intimate knowledge of generative AI systems. The new APIs should make implementing local AI comparatively quick and easy. Despite the limited functionality of an on-device model, this is an important part of how AI could become more helpful. Most people would probably prefer not to send all their personal data to a remote server for AI processing, but an on-device model can parse that information in a more secure way. For example, Google's Pixel Screenshots sees all your screenshots, but all the processing happens on your phone. Similarly, Motorola summarizes notifications locally on the new Razr Ultra foldable. On the other hand, its less capable base model Razr sends notifications to a server for processing. The release of APIs that plug into Gemini Nano could provide some much-needed consistency to mobile AI. However, it does rely on Google and OEMs to collaborate on support for Gemini Nano. Some companies might decide to go their own way, and there will be plenty of phones that don't have enough power to run AI locally. Ryan Whitwam Senior Technology Reporter Ryan Whitwam Senior Technology Reporter Ryan Whitwam is a senior technology reporter at Ars Technica, covering the ways Google, AI, and mobile technology continue to change the world. Over his 20-year career, he's written for Android Police, ExtremeTech, Wirecutter, NY Times, and more. He has reviewed more phones than most people will ever own. You can follow him on Bluesky, where you will see photos of his dozens of mechanical keyboards. 16 Comments

AI evolution Google DeepMind creates super-advanced AI that can invent new algorithms AlphaEvolve has already made Google's data centers more efficient and improved Tensor chips. Ryan Whitwam – May 14, 2025 5:01 pm | 17 Credit: Google DeepMind Credit: Google DeepMind Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more Google's DeepMind research division claims its newest AI agent marks a significant step toward using the technology to tackle big problems in math and science. The system, known as AlphaEvolve, is based on the company's Gemini large language models (LLMs), with the addition of an "evolutionary" approach that evaluates and improves algorithms across a range of use cases. AlphaEvolve is essentially an AI coding agent, but it goes deeper than a standard Gemini chatbot. When you talk to Gemini, there is always a risk of hallucination, where the AI makes up details due to the non-deterministic nature of the underlying technology. AlphaEvolve uses an interesting approach to increase its accuracy when handling complex algorithmic problems. According to DeepMind, this AI uses an automatic evaluation system. When a researcher interacts with AlphaEvolve, they input a problem along with possible solutions and avenues to explore. The model generates multiple possible solutions, using the efficient Gemini Flash and the more detail-oriented Gemini Pro, and then each solution is analyzed by the evaluator. An evolutionary framework allows AlphaEvolve to focus on the best solution and improve upon it. Many of the company's past AI systems, for example, the protein-folding AlphaFold, were trained extensively on a single domain of knowledge. AlphaEvolve, however, is more dynamic. DeepMind says AlphaEvolve is a general-purpose AI that can aid research in any programming or algorithmic problem. And Google has already started to deploy it across its sprawling business with positive results. The team turned AlphaEvolve loose on Google's Borg cluster management system for its data centers. The AI suggested a change to the scheduling heuristics, which has been implemented to save Google 0.7 percent on its computing resources globally. For a company the size of Google, that's a significant financial benefit. AlphaEvolve may also be able to make generative AI more efficient, which is necessary if anyone is ever going to make money on the technology. The internal workings of generative systems are based on matrix multiplication operations. The most efficient way to multiply 4×4 complex-valued matrices was devised by mathematician Volker Strassen in 1969, and that held for decades, but DeepMind says AlphaEvolve has discovered a new algorithm that's even more efficient. DeepMind has worked on this problem before with narrowly trained AI agents like AlphaTensor. Despite being a general AI, AlphaEvolve came up with a better solution than AlphaTensor. Google's next-generation Tensor processing hardware will also benefit from AlphaEvolve. DeepMind reports that the AI created a change to the chip's Verilog hardware description language that dropped unnecessary bits to increase efficiency. Google is still working to verify the change but expects this to be part of the upcoming processor. So far, only Google has been able to tinker with AlphaEvolve. While it uses fewer computing resources than AlphaTensor did, it's still too complex to provide publicly. That may change in the future, but the evaluation approach that makes it so capable could also be integrated with smaller AI tools for research. Ryan Whitwam Senior Technology Reporter Ryan Whitwam Senior Technology Reporter Ryan Whitwam is a senior technology reporter at Ars Technica, covering the ways Google, AI, and mobile technology continue to change the world. Over his 20-year career, he's written for Android Police, ExtremeTech, Wirecutter, NY Times, and more. He has reviewed more phones than most people will ever own. You can follow him on Bluesky, where you will see photos of his dozens of mechanical keyboards. 17 Comments

entremet You wouldn't toast a NES cartridge Member Oct 26, 2017 70,440 I've also loved this troupe. It's like a fun fakeout lol. My pick: Monster World IV, the last Genesis installment of the installment of the excellent, and confusing Monster World/Wonder Boy series. It also got a remake a few years ago as well, which I haven't played. End game just goes crazy while you basically coast for 75 percent of the game. It's not crazy, but a nice jump in difficulty.  Last edited: Yesterday at 10:21 PM PlanetSmasher The Abominable Showman Member Oct 25, 2017 132,547 The last few missions of Super Robot Wars OG: Moon Dwellers are nuts. The game is like "we're going to give you one of the most powerful mechas in the entire multiverse! so now we need to make every fight 2-3x harder to compensate!" The Neo Granzon is an absolute monster, but even that thing can't solve an entire mission by itself if the rest of your units get killed.  shadowman16 Member Oct 25, 2017 41,370 Ill be honest the remake of MWIV didnt really feel like it had a difficulty spike aside that darned quiz that popped up mid? way through the game. Its basically a knowledge check, in a game which Id been sleeping through the story. That'll teach me to pay more attention next time lol. Ones that hit me personally: Ninja Gaiden (NES) - That darned bird in stage... 7-2? Its actually not just the bird but rather a combo of the bird and other enemies that constantly respawn. The level actually wouldnt be as bad if the respawning was more lenient. Or you know, the checkpointing after that point was solid instead of kicking you back to 7-1 if you lose to the boss' multiple phases! Ninja Gaiden (Black/Sigma) - Bump the difficulty all the way up to Very Hard/Master Ninja, then play the penultimate stage and do the boss rush run toward the end of the penultimate chapter... when the Alma like bosses show up, it can become a real test of skill just to hit the bosses, let alone win! TMNT NES - No, not the over rated dam level, Im talking level 6 - the finale. And specifically, the long corridor section right before Shredder - the game basically over and over spawns these flying dudes with guns that decimate your health and take far too many hits to defeat... they dont stop coming and are just a hastle to deal with. Hope for scrolls or at least limp to Shredder's boss room with Don and Leo intact for the fight, as you'll need their reach to cheese Shredder (thankfully said boss is much easier than the journey to him)  DrForester Mod of the Year 2006 Member Oct 25, 2017 24,538 Breath of the Wild, but only because I was so engrossed during the game of exploring on foot that having to do mounted combat for the final battle was not something I was prepared for at all.   OP OP entremet You wouldn't toast a NES cartridge Member Oct 26, 2017 70,440 shadowman16 said: Ill be honest the remake of MWIV didnt really feel like it had a difficulty spike aside that darned quiz that popped up mid? way through the game. Its basically a knowledge check, in a game which Id been sleeping through the story. That'll teach me to pay more attention next time lol. Ones that hit me personally: Ninja Gaiden (NES) - That darned bird in stage... 7-2? Its actually not just the bird but rather a combo of the bird and other enemies that constantly respawn. The level actually wouldnt be as bad if the respawning was more lenient. Or you know, the checkpointing after that point was solid instead of kicking you back to 7-1 if you lose to the boss' multiple phases! Ninja Gaiden (Black/Sigma) - Bump the difficulty all the way up to Very Hard/Master Ninja, then play the penultimate stage and do the boss rush run toward the end of the penultimate chapter... when the Alma like bosses show up, it can become a real test of skill just to hit the bosses, let alone win! TMNT NES - No, not the over rated dam level, Im talking level 6 - the finale. And specifically, the long corridor section right before Shredder - the game basically over and over spawns these flying dudes with guns that decimate your health and take far too many hits to defeat... they dont stop coming and are just a hastle to deal with. Hope for scrolls or at least limp to Shredder's boss room with Don and Leo intact for the fight, as you'll need their reach to cheese Shredder (thankfully said boss is much easier than the journey to him) Click to expand... Click to shrink... I heard the remake is much easier than the Genesis original.   BennyWhatever Member Oct 27, 2017 5,485 US The Last Story for the Wii, I remember the very last fight being a huuuuge difficulty spike. I breezed through most of the game but the final boss took me hours.   JaseMath Member Oct 27, 2017 11,422 Denver, CO The endgame of FFIV (Lunar Subterranean) can be pretty brutal if you're not grinding it.   shadowman16 Member Oct 25, 2017 41,370 entremet said: I heard the remake is much easier than the Genesis original. Click to expand... Click to shrink... Ive not played the original yet, but yeah I can believe it. Still I really liked the remake for what its worth.   Tenck Member Oct 27, 2017 684 Yakuza: Like a Dragon. I almost dropped the game because of it, but because I was so close to the end I was stubborn and pushed through the difficulty spike. That game all around had some bad moments when it came to difficulty, but the ending was on a whole different level.   BranCrackr Member Oct 4, 2019 1,260 Not a complaint for me, but Stellar Blade's last few bosses will beat your ass if you haven't gotten good at parrying and perfect dodging   MimicBox82 Member Jan 9, 2018 5,406 Sweden Elden Ring. Those last few bosses are brutal.   intradictus Member Sep 17, 2023 567 Senator Armstrong in Metal Gear Rising: Revengeance, you can make it to him without mastering the combat system, but he will beat your ass if you have any weaknesses   sukitakit4 Member Aug 2, 2022 25 intradictus said: Senator Armstrong in Metal Gear Rising: Revengeance, you can make it to him without mastering the combat system, but he will beat your ass if you have any weaknesses Click to expand... Click to shrink... Yep...pretty much.  thetrin Member Oct 26, 2017 10,581 Grand Junction, CO Tenck said: Yakuza: Like a Dragon. I almost dropped the game because of it, but because I was so close to the end I was stubborn and pushed through the difficulty spike. That game all around had some bad moments when it came to difficulty, but the ending was on a whole different level. Click to expand... Click to shrink... IF I am thinking of the same part of the game, the difficulty spikes in a really stupid way too. Its not that the enemies are at your level but more difficult. If you dont grind before this point (something you never really need to do up to this point), everything out levels you i that section by several levels. It's bizarre balancing.  cubicle47b Member Aug 9, 2019 921 The real final boss of Octopath Traveler.   JohnTruckasaurus Member Oct 27, 2017 308 The last level of Streets of Rage 4 on a higher difficulty is pretty much designed to make your palms sweat.   3bada The Fallen Oct 27, 2017 512 Kena is a chill adventure game until the final act where it becomes dark souls   Listai 50¢ - "This guy are sick" Member Oct 27, 2017 6,819 Final Fantasy III. I was overlevelled the whole game but that final dungeon was brutal.  Tavernade Tavernade Moderator Sep 18, 2018 11,609 The final boss of the first Mario and Luigi game is total nonsense compared to the rest of the game. Two phases, the second of which resets your HP to zero and then sends you through a gauntlet.   sillyGecko Member Mar 14, 2025 1,262 Does Astrobot count? Those challenge levels are brutal  

A fresh coat of paint Google announces Material 3 Expressive, a colorful evolution of Android design Google's take on Android is getting brighter and bolder later this year. Ryan Whitwam – May 13, 2025 2:57 pm | 32 Credit: Google Credit: Google Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more Google accidentally showed off its big Android design refresh last week, but now Material 3 Expressive is official. Google says the new interface will begin with Android OS, but it will eventually expand across the full Google app ecosystem, bringing a more lively vibe to Gmail, Google Photos, and more. Material 3 Expressive won't be entirely unfamiliar—it shares some basic design elements with the Material You system Google launched four years ago. Material 3 Expressive is a bolder take on the same aesthetic, featuring "springy" animations, brighter colors, and new shapes. Material 3 Expressive As we learned from Google's slipup, Material 3 Expressive is the result of numerous user studies, which included more than 18,000 participants. Google explored how people parse information on their phone, finding design themes that supposedly support quicker and easier interactions. Often, that seems to result in making certain UI elements larger and easier to spot. Google claims people can find these important buttons four times faster than they can with older Material You interfaces. What you can expect from Material 3 Expressive Animations will abound in future versions of Android, and Google emphasizes the springy aspect. The style is supposed to feel natural and fun, with more UI elements connected to dynamic haptics. Colors are also getting an overhaul, but this isn't a complete rethink of Material You. The theming engine will pick bolder colors to improve the visual separation of UI elements, and you will see those colors in more places. Google says it's also updating Android's typography, featuring more contrast between headers and body text. This will change across Google's apps to help users parse information faster. Certain buttons will also come in a wider variety of shapes, and the labels will have varying text weights. Likewise, the style and shape of status bar icons will change to be more readable. The notification shade and quick settings may be one of the more dramatic shifts for Material 3 Expressive. Google will finally let you pin more controls to the compact view, and the quick settings tiles will be resizable. The background of the notification shade will use a blur effect on your background image as well. While Android 16 is just around the corner, and elements of Material 3 Expressive are hiding in the latest betas, this is not the default UI for Google's new version. Android 16 will launch in June with the Material You interface you know and (maybe) love. Later this year, Material 3 Expressive will come to Pixel phones and the Pixel Watch via an update. And what of non-Google devices? Like Material You, Material 3 Expressive is an open design framework. Any of Google's partners can implement it in their Android skin, and developers will be able to take advantage of design templates and new animation APIs to make their apps blend in with Material 3 Expressive. But will they? History suggests that Google's design preferences are rarely reflected in software from companies like Samsung and OnePlus. Motorola has traditionally hewed closer to Google's design language, but even Moto didn't fully embrace Material You. If you love the vibes of Material 3 Expressive, you'll probably need a Pixel to get the full effect. Ryan Whitwam Senior Technology Reporter Ryan Whitwam Senior Technology Reporter Ryan Whitwam is a senior technology reporter at Ars Technica, covering the ways Google, AI, and mobile technology continue to change the world. Over his 20-year career, he's written for Android Police, ExtremeTech, Wirecutter, NY Times, and more. He has reviewed more phones than most people will ever own. You can follow him on Bluesky, where you will see photos of his dozens of mechanical keyboards. 32 Comments