Latest news with #2.5Flash
Techday NZ
22-05-2025
- Business
- Techday NZ
Google unveils Gemini 2.5 upgrades for reasoning & security
Google has provided a series of updates to its Gemini 2.5 model series, with enhancements spanning advanced reasoning, developer capabilities and security safeguards. The company reported that Gemini 2.5 Pro is now the leading model on the WebDev Arena coding leaderboard, holding an ELO score of 1415. It also leads across all leaderboards in LMArena, a platform that measures human preferences in multiple dimensions. Additionally, Gemini 2.5 Pro's 1 million-token context window was highlighted as supporting strong long context and video understanding performance. Integration with LearnLM, a family of models developed with educational experts, resulted in Gemini 2.5 Pro apparently becoming the foremost model for learning. According to Google, in direct comparisons focusing on pedagogy and effectiveness, Gemini 2.5 Pro was favoured by educators and experts over other models in a wide range of scenarios. The model outperformed others based on the five principles of learning science used in AI system design for education. Gemini 2.5 Pro introduced an experimental capability called Deep Think, which is being tested to enable enhanced reasoning by allowing the model to consider multiple hypotheses before responding. The company said, "2.5 Pro Deep Think gets an impressive score on 2025 USAMO, currently one of the hardest math benchmarks. It also leads on LiveCodeBench, a difficult benchmark for competition-level coding, and scores 84.0% on MMMU, which tests multimodal reasoning." Safety and evaluation measures are being emphasised with Deep Think. "Because we're defining the frontier with 2.5 Pro DeepThink, we're taking extra time to conduct more frontier safety evaluations and get further input from safety experts. As part of that, we're going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available," the company reported. Google announced improvements to 2.5 Flash, describing it as the most efficient in the series, tailored for speed and cost efficiency. This version now reportedly uses 20-30% fewer tokens in evaluations and delivers improved performance across benchmarks for reasoning, multimodality, code, and long-context tasks. The updated 2.5 Flash is now available for preview in Google AI Studio, Vertex AI, and the Gemini app. New features have also been added to the Gemini 2.5 series. The Live API now offers a preview version supporting audio-visual input and native audio output. This is designed to create more natural and expressive conversational experiences. According to Google, "It also allows the user to steer its tone, accent and style of speaking. For example, you can tell the model to use a dramatic voice when telling a story. And it supports tool use, to be able to search on your behalf." Early features in this update include Affective Dialogue, where the model can detect and respond to emotions in a user's voice; Proactive Audio, which enables the model to ignore background conversations and determine when to respond; and enhanced reasoning in live API use. Multi-speaker support has also been introduced for text-to-speech capabilities, allowing audio generation with two distinct voices and support for over 24 languages, including seamless transitions between them. Project Mariner's computer use capabilities are being integrated into the Gemini API and Vertex AI, with multiple enterprises testing the tool. Google stated, "Companies like Automation Anywhere, UiPath, Browserbase, Autotab, The Interaction Company and Cartwheel are exploring its potential, and we're excited to roll it out more broadly for developers to experiment with this summer." On the security front, Gemini 2.5 includes advanced safeguards against indirect prompt injections, which involve malicious instructions embedded into retrieved data. According to disclosures, "Our new security approach helped significantly increase Gemini's protection rate against indirect prompt injection attacks during tool use, making Gemini 2.5 our most secure model family to date." Google is introducing new developer tools with thought summaries in the Gemini API and Vertex AI. These summaries convert the model's raw processing into structured formats with headers and action notes. Google stated, "We hope that with a more structured, streamlined format on the model's thinking process, developers and users will find the interactions with Gemini models easier to understand and debug." Additional features include thinking budgets for 2.5 Pro, allowing developers to control the model's computation resources to balance quality and speed. This can also completely disable the model's advanced reasoning capability if desired. Model Context Protocol (MCP) support has been added for SDK integration, aiming to enable easier development of agentic applications using both open-source and hosted tools. Google affirmed its intention to sustain research and development efforts as the Gemini 2.5 series evolves, stating, "We're always innovating on new approaches to improve our models and our developer experience, including making them more efficient and performant, and continuing to respond to developer feedback, so please keep it coming! We also continue to double down on the breadth and depth of our fundamental research — pushing the frontiers of Gemini's capabilities. More to come soon."
Yahoo
20-05-2025
- Business
- Yahoo
Google introduces the Deep Think reasoning model for Gemini 2.5 Pro and a better 2.5 Flash
Google has started testing a reasoning model called Deep Think for Gemini 2.5 Pro, the company has revealed at its I/O developer conference. According to DeepMind CEO Demis Hassabis, Gemini's Deep Think uses "the latest cutting-edge research" that gives the model the capability to consider multiple hypotheses before responding to queries. Google says it got an "impressive score" when evaluated using questions from the 2025 United States of America Mathematical Olympiad competition. However, Google wants to take more time to conduct safety evaluations and get further input from safety experts before releasing it widely. That's why it's making Deep Think initially available to trusted testers via the Gemini API first in order to get their feedback first. The company has also introduced a better Gemini 2.5 Flash model, which is optimized for speed and efficiency. It's now more efficient than before, uses fewer tokens and has scored higher in benchmarks for reasoning, multimodality, code and long context than its predecessor. It will be generally available in early June. For now, the improved Gemini 2.5 Flash is available as a preview via Google AI Studio for developers, via Vertex AI for enterprise customers and via the Gemini app for other users. While most of the efficiency gains covered on the I/O stage were focused on 2.5 Flash, Google did announce that it's bringing the 2.5 Flash concept of "Thinking Budgets" to its more advanced 2.5 Pro model. This feature will let you balance tokens spent vs. accuracy and speed of output. Separately, Google is bringing Project Mariner into the Gemini API and Vertex AI, as well. Project Mariner is Google's Gemini-powered AI agents that can navigate pages on the web browser to complete tasks for users. The company will roll the agents out more broadly this summer so that developers can experiment with them. In addition, the company is releasing new previews for text-to-speech on both 2.5 Pro and 2.5 Flash models via the Gemini API, with support for two voices in 24 languages.
Engadget
20-05-2025
- Business
- Engadget
Google introduces the Deep Think reasoning model for Gemini 2.5 Pro and a better 2.5 Flash
Google has started testing a reasoning model called Deep Think for Gemini 2.5 Pro, the company has revealed at its I/O developer conference. According to DeepMind CEO Demis Hassabis, Gemini's Deep Think uses "the latest cutting-edge research" that gives the model the capability to consider multiple hypotheses before responding to queries. Google says it got an "impressive score" when evaluated using questions from the 2025 United States of America Mathematical Olympiad competition. However, Google wants to take more time to conduct safety evaluations and get further input from safety experts before releasing it widely. That's why it's making Deep Think initially available to trusted testers via the Gemini API first in order to get their feedback first. The company has also introduced a better Gemini 2.5 Flash model, which is optimized for speed and efficiency. It's now more efficient than before, uses fewer tokens and has scored higher in benchmarks for reasoning, multimodality, code and long context than its predecessor. It will be generally available in early June. For now, the improved Gemini 2.5 Flash is available as a preview via Google AI Studio for developers, via Vertex AI for enterprise customers and via the Gemini app for other users. While most of the efficiency gains covered on the I/O stage were focused on 2.5 Flash, Google did announce that it's bringing the 2.5 Flash concept of "Thinking Budgets" to its more advanced 2.5 Pro model. This feature will let you balance tokens spent vs. accuracy and speed of output. Separately, Google is bringing Project Mariner into the Gemini API and Vertex AI, as well. Project Mariner is Google's Gemini-powered AI agents that can navigate pages on the web browser to complete tasks for users. The company will roll the agents out more broadly this summer so that developers can experiment with them. In addition, the company is releasing new previews for text-to-speech on both 2.5 Pro and 2.5 Flash models via the Gemini API, with support for two voices in 24 languages.
Tom's Guide
20-05-2025
- Business
- Tom's Guide
Gemini just saw a huge upgrade to its AI model — here's everything new you can do
Google Gemini is stepping up its game. Announced at Google's yearly I/O event, the AI tool has just gotten one of its biggest updates ever, seeing improvements across multiple models and bringing in new features. This covers everything from improvements in coding and web design, to boosts in model efficiency and a brand new deep research feature. On top of this, Google has announced updates to its AI video generator with Veo 3, as well as announcing new AI plans and other improvements to its suite of AI tools. But, for now, let's focus on how Gemini looks different and all the new changes that are coming. A new feature announced at I/O for Gemini, Deep Think, is an enhanced reasoning mode. This uses new research techniques, enabling the model to consider multiple different hypotheses before responding. This is a concept that fits in well with reasoning models, where AI can think through a task with more detail. Google claims that 2.5 Pro Deep Think scored impressively on one of the hardest math benchmarks available, as well as leading on multiple AI testing systems for multimodal reasoning. Get instant access to breaking news, the hottest reviews, great deals and helpful tips. Deep Think won't be immediately available, and no release date has been announced yet. Google explained that they want to do further testing before they release this feature to the public. An update coming to Gemini in early June, Google announced improvements to its 2.5 Flash mode. This is the model designed for speed and low-cost tasks, build for simple prompting without the model needing to think too deeply. Google revealed that 2.5 Flash has been improved across key benchmarks for reasoning, multimodality, coding and long context. All of this while using 20 to 30% fewer tokens (computing power). The main update for 2.5 Pro from Google came early. Announced in the first week of May, this update improved Gemini 2.5 Pro's ability to build interactive web apps. This was a major improvement on the technology, and saw a big push for vibe coding (the ability to code through AI prompts). Announced at I/O, Google also claimed that the new 2.5 Pro model is now leading the popular coding leaderboard WebDev Arena, as well as leading multiple categories of the LMArena. These are tests of both how well the model can develop websites, and also their ability to take on tasks like image generation, and how efficient they are. Google claimed that with the improvements launched earlier this month, Gemini is now the leading model for learning, outperforming top models on every one of the five principles of learning science. One of the more interesting announcements out of the Gemini reveal is a feature where users can customise the dialogue of Gemini Live. This could be used to make Gemini more natural and expressive, allowing users to steer its tone, accent, and style of speaking. This will include a variety of new tools like affective dialogue, where the model detects emotion in your voice and replies accordingly. Proactive audio, where the model ignores background conversations and knows when to respond. And finally, deep thinking in Live conversations. This will first be made available in the Gemini API system for developers, but will likely then follow onto Gemini. Project Mariner will be coming to the Gemini API and Vertex AI. This is a research tool that enables human-agent interaction. In other words, it could allow Gemini to complete tasks across websites, like booking flights, completing forms, and following workflow summaries. For now, this will only be available to developers to experiment with and there is no detail of a future release on Gemini. Google claims that, with this latest update, Gemini 2.5 is the most secure AI model family they've made. This includes making improvements to protections against security threats and malicious instructions that could be embedded into the models. You'll likely not notice any changes here, but that just means it's working well! Further updates were announced specifically for developers using Gemini tools. These were broken down into three sections: Thought summaries, thinking budgets and MCP support. Thought Summaries are a new ability in Gemini API and Vertex AI, in which the model will summarize its raw thoughts and organize them with headers and key details. Thinking budgets allow developers to have more control over cost by balancing latency and quality, allowing them to control the number of tokens a model uses before it responds. Finally, MCP Support will make it easier to integrate the Gemini API with open-source tools. Google claims that it is also working on new approaches to improve the model and developer experience. This is why the tools will first be available to developers.
Yahoo
10-04-2025
- Business
- Yahoo
Google's newest Gemini AI model focuses on efficiency
Google is releasing a new AI model designed to deliver strong performance with a focus on efficiency. The model, Gemini 2.5 Flash, will soon launch in Vertex AI, Google's AI development platform. The company says it offers "dynamic and controllable" computing, allowing developers to adjust processing time based on the complexity of queries. "[You can tune] the speed, accuracy, and cost balance for your specific needs," Google wrote in a blog post provided to TechCrunch. "This flexibility is key to optimizing Flash performance in high-volume, cost-sensitive applications." Gemini 2.5 Flash arrives as the cost of flagship AI models continues trending upward. Lower-priced performant models like 2.5 Flash present an attractive alternative to expensive top-of-the-line options at the cost of some accuracy. Gemini 2.5 Flash is a "reasoning" model along the lines of OpenAI's o3-mini and DeepSeek's R1. That means it takes a bit longer to answer questions in order to fact-check itself. Google says that 2.5 Flash is ideal for "high-volume" and "real-time" applications like customer service and document parsing. "This workhorse model is optimized specifically for low latency and reduced cost," Google said in its blog post. "It's the ideal engine for responsive virtual assistants and real-time summarization tools where efficiency at scale is key." Google didn't publish a safety or technical report for Gemini 2.5 Flash, making it more challenging to see where the model excels and falls short. The company previously told TechCrunch that it doesn't release reports for models it considers to be "experimental." Google also announced on Wednesday that it plans to bring Gemini models like 2.5 Flash to on-premises environments starting in Q3. The company's Gemini models will be available on Google Distributed Cloud (GDC), Google's on-prem solution for clients with strict data governance requirements. Google says it's working with Nvidia to bring Gemini models to GDC-compliant Nvidia Blackwell systems that customers can purchase through Google or their preferred channels. Sign in to access your portfolio
