Google has recently updated the way it evaluates the AI models employed in Android development.

Google has recently updated the way it evaluates the AI models employed in Android development.

      Android Bench has introduced a new testing framework along with eight new models, so the previous rankings you may recall are now outdated.

      Google has modified the way it evaluates which AI models are most effective at writing Android app code, resulting in a rearrangement of the rankings that developers utilize to select their tools. The Android Bench leaderboard, which was introduced in March, now operates on a new testing system called Harbor. Google states that this system replaces the older, more generic testing tool previously in use, offering a more accurate assessment of how models perform on actual Android tasks, such as updating legacy code to Jetpack Compose or managing networking for wearable devices.

      New models alter the leaderboard standings

      With the change in the testing tool, Google re-evaluated each model. Eight new models have been included in the leaderboard, such as Claude Fable 5, Claude Sonnet 5, Claude Opus 4.8, GLM 5.2, Kimi K2.7 Code, MiniMax M3, Qwen 3.7 Plus, and Qwen 3.7 Max.

      As of now, Claude Fable 5 leads the list with a score of 84.5 percent, followed by GPT 5.5 at 80.2 percent and Claude Sonnet 5 at 76.2 percent. Among free, open-weight models, GLM 5.2 tops the list with a score of 72.2 percent, just ahead of Kimi K2.7 Code at 70.4 percent. Google’s own Gemini 3.1 Pro ranks fifth with a score of 73.7 percent. If you’re using Android Bench to choose the right model for your coding tasks, it's advisable to visit the Android Bench website to check the updated leaderboard.

      Developers can now contribute their own test tasks

      When Google launched Android Bench in March, it shared its testing methodology and test harness on GitHub for greater transparency. Now, the initiative has been expanded by allowing developers to submit their own Android development tasks for benchmarking. Developers can also conduct tests with their preferred models and share their findings with the community.

      The rankings are expected to shift as more new models appear, including OpenAI’s recently launched GPT-5.6 Sol, Terra, and Luna. If you’re selecting an AI tool for Android coding, consider the current leaderboard a momentary snapshot rather than a fixed ranking, and remember to check back before finalizing your choice for the next project.

      Pranob is an experienced tech journalist with over eight years of experience in consumer technology coverage. His work has been…

      You can now verify if a Google ad was generated using AI

      Google will automatically label its own AI-generated ads, but third-party ads produced via AI still depend on advertisers being transparent.

      Have you ever looked at an ad and speculated whether it was created by a person or generated by AI in moments? Google is now providing a way to find out. The company has recently introduced a new AI transparency label indicating whether an ad was created or modified using generative AI tools. This label will be available in Google’s My Ad Center and will gradually roll out across Google Search, YouTube, and Discover worldwide.

      Microsoft Outlook will soon notify you before responding to an outdated email

      Microsoft is introducing reply alerts, rule-based templates, and enhanced categories to Outlook.

      Microsoft has been addressing a number of long-standing issues in Windows 11, including aspects of the Start menu and Search. As per a recent report from Windows Latest, the company is also preparing several beneficial updates for the new Outlook app on Windows 10 and Windows 11, which became widely available in 2024. Microsoft is adding a feature that alerts users when they attempt to reply to an older email after receiving a newer reply in the same thread. This alert aims to prevent users from responding without being aware of the latest updates in the conversation.

      ChatGPT is set to rival one of Google’s most intelligent Chrome features

      OpenAI introduces ChatGPT to Chrome to compete with Google’s Gemini Side Panel.

      OpenAI is expanding the reach of ChatGPT with the launch of a new Chrome extension that can comprehend the content of the webpage you are viewing. This extension allows users to inquire about a page, summarize articles, explain intricate concepts, and even initiate longer AI-powered tasks without leaving their browser. This development positions ChatGPT as a direct competitor to Google’s Gemini in Chrome, which introduced similar context-aware browsing functionalities earlier this year. While both tools aim to integrate AI directly into web browsing, they offer slightly different approaches to productivity and automation.

Google has recently updated the way it evaluates the AI models employed in Android development. Google has recently updated the way it evaluates the AI models employed in Android development. Google has recently updated the way it evaluates the AI models employed in Android development. Google has recently updated the way it evaluates the AI models employed in Android development. Google has recently updated the way it evaluates the AI models employed in Android development. Google has recently updated the way it evaluates the AI models employed in Android development.

Other articles

China successfully lands a reusable rocket booster for the first time. China successfully lands a reusable rocket booster for the first time. China successfully captured a Long March-10B booster in a net at sea, marking its initial recovery of a reusable rocket. This places it alongside SpaceX and Blue Origin, though it significantly lags behind them. This small device known as Moodi might protect your thumb during extended reading periods. This small device known as Moodi might protect your thumb during extended reading periods. DuRoBo has launched Moodi, a compact Bluetooth remote that enables users to flip pages, navigate documents, and manage media without the need to physically interact with their device. Kraken secures $175 million to achieve unicorn status in maritime defense. Kraken secures $175 million to achieve unicorn status in maritime defense. The British company Kraken secured $175 million (€152.9 million), with DTCP leading the round, achieving a valuation of $1 billion to develop uncrewed war boats for NATO, the United States, and the United Kingdom. MiniMax secures $2 billion as its CEO waives salary until AGI is achieved. MiniMax is seeking to raise as much as $2 billion following an 80% decline, with founder Yan Junjie declaring he will forgo his salary until artificial general intelligence (AGI) is achieved and promising 5% of his shares to employees and for open-source initiatives. Mercor aims for a $20 billion valuation and acquires Deeptune. Mercor aims for a $20 billion valuation and acquires Deeptune. Mercor is negotiating to increase its valuation to $20 billion and has acquired Deeptune, an AI-training startup supported by its own CEO. The details are in the footnotes. Facewatch will notify UK police within seconds if you are flagged. Facewatch will notify UK police within seconds if you are flagged. Facewatch will soon notify the police within approximately four seconds when a flagged customer enters a store in the UK. Civil liberties organizations describe this as a "dangerous escalation."

Google has recently updated the way it evaluates the AI models employed in Android development.

Google has updated its Android Bench leaderboard by introducing a new testing framework and adding eight new models.