Gemini 2.5 Flash Lite
Gemini 2.5 Flash Lite is Google’s cheapest multimodal model, designed for cost-sensitive workloads where latency and price matter more than peak quality.
Released
June 17, 2025
Type
multimodal
License
proprietary
Context
1,000,000 tokens
Pricing
Input
$0.07 / 1M tokens
Output
$0.30 / 1M tokens
Capabilities
Links
Gemini 2.5 Flash Lite in the news
Al Jazeera · Sep 9, 2026
AI researcher quits Anthropic saying AI race ‘could kill us all’
An AI researcher has resigned from Anthropic, a prominent artificial intelligence company, expressing grave concerns about the accelerating pace of AI development and its potential existential risks. The researcher, whose name has not been publicly disclosed, stated that the current trajectory of the AI race is leading towards the creation of superintelligent AI that could surpass human control, posing a significant threat to humanity's future. This departure highlights growing anxieties within the AI community regarding safety and the long-term implications of increasingly powerful AI systems. Anthropic, founded in 2021 by former OpenAI employees, has positioned itself as a leader in developing safe and beneficial AI. The company's stated mission is to build reliable, interpretable, and steerable AI systems. However, the researcher's resignation suggests that even within organizations dedicated to AI safety, fundamental disagreements and anxieties persist about the speed and direction of progress. The researcher's warning echoes concerns previously voiced by other prominent figures in the AI field, including OpenAI CEO Sam Altman and Google DeepMind CEO Demis Hassabis, who have acknowledged the potential for AI to pose societal risks if not developed and managed responsibly. The core of the researcher's concern revolves around the concept of recursive self-improvement, where an AI system becomes capable of improving its own intelligence at an exponential rate. This could lead to an intelligence explosion, resulting in an AI far exceeding human cognitive abilities and potentially acting in ways detrimental to human interests. The researcher's decision to leave Anthropic, a company known for its focus on AI alignment and safety research, underscores the perceived urgency of these risks. It suggests that the internal discussions and safeguards within even leading AI labs may not be sufficient to mitigate the potential dangers of advanced AI. The AI industry is currently characterized by intense competition among major players like OpenAI, Google DeepMind, Meta AI, and Anthropic. This competitive landscape, driven by significant investment and the pursuit of groundbreaking capabilities, may inadvertently incentivize faster development over rigorous safety considerations. The researcher's statement implies that the drive to be first to achieve advanced AI milestones could be overshadowing the critical need for robust safety protocols and a deeper understanding of the potential consequences. The resignation serves as a stark reminder of the ethical and safety challenges inherent in the pursuit of artificial general intelligence (AGI) and superintelligence.
Financial Times · Sep 9, 2026
Anthropic researcher quits over AI labs ‘gambling with our lives’
A researcher has resigned from Anthropic, a leading artificial intelligence company, citing profound concerns that AI laboratories are "gambling with our lives" in their pursuit of increasingly powerful AI systems. The departure, detailed in a recent account, underscores a growing unease within Silicon Valley companies that are engaged in an intense race to develop advanced AI capabilities. The researcher, whose identity has not been publicly disclosed, expressed a belief that the rapid, unchecked development of AI poses existential risks that are not being adequately addressed by the organizations at the forefront of this technological frontier. This resignation highlights a broader trend of internal dissent and ethical questioning within the AI industry. As companies like Anthropic, OpenAI, and Google DeepMind push the boundaries of AI development, they face increasing scrutiny not only from external regulators and the public but also from their own employees. The core of the concern revolves around the potential for superintelligent AI systems to act in ways that are unpredictable and potentially harmful to humanity, a scenario often referred to as an existential risk. The researcher's decision to leave suggests that the perceived risks outweigh the potential benefits or that the current safety measures are insufficient to mitigate these dangers. Anthropic, co-founded by former OpenAI researchers, has positioned itself as a company prioritizing AI safety and alignment, aiming to build AI systems that are beneficial and harmless. The company has publicly discussed its efforts to develop constitutional AI, a method designed to train AI models to adhere to a set of ethical principles. However, the researcher's resignation implies that these efforts, while perhaps well-intentioned, are not perceived as adequate by those within the organization who are deeply involved in the technical development. The statement "gambling with our lives" suggests a perception that the pace of development is outstripping the capacity for robust safety research and implementation, leading to a situation where the future of humanity is being put at risk. The competitive landscape of AI development is characterized by a relentless drive for innovation, often fueled by significant investment and the fear of falling behind rivals. This environment can create pressure to prioritize speed and capability over caution and safety. The researcher's concerns echo sentiments previously voiced by various AI safety advocates and some prominent figures within the AI community, who have warned about the potential for unintended consequences from advanced AI. The departure from Anthropic serves as a concrete example of these abstract concerns manifesting as real-world actions by individuals directly involved in the creation of these powerful technologies. The implications of such resignations extend beyond the individual, potentially influencing public perception, regulatory discussions, and the internal culture of AI development companies.
TechCrunch · Sep 9, 2026
Google DeepMind alumni are building tools to accelerate fusion power for the grid
Fusionality, a new company founded by former researchers from Google DeepMind, is developing tools designed to accelerate the development of fusion power for electricity generation. The startup's mission is to provide the necessary technological advancements that will enable fusion power to become a viable and scalable energy source for the global grid. Fusionality's core offerings include sophisticated control systems and comprehensive simulation environments specifically tailored for fusion power startups. These tools are intended to streamline the complex processes involved in fusion research and development, allowing companies in the sector to progress more rapidly towards achieving sustained fusion reactions. The founders of Fusionality bring significant expertise from their tenure at Google DeepMind, a leading artificial intelligence research laboratory known for its contributions to areas such as reinforcement learning and scientific discovery. This background is expected to inform the development of innovative solutions for the challenges inherent in fusion energy. Achieving controlled nuclear fusion, the process that powers stars, has long been a goal for clean energy advocates due to its potential for abundant, carbon-free power with minimal long-lived radioactive waste. However, the engineering and physics challenges are immense, requiring precise control over extremely high-temperature plasmas and complex magnetic confinement or inertial confinement systems. Fusionality's approach focuses on providing the software and simulation infrastructure that can de-risk and speed up experimental fusion projects. By offering advanced control algorithms, potentially leveraging AI techniques honed at DeepMind, the company aims to help fusion startups optimize their reactor designs and operational parameters. The simulation environments will allow for virtual testing and refinement of fusion reactor concepts before significant physical resources are committed, thereby reducing development timelines and costs. This initiative comes at a time when global investment in fusion energy is increasing, with numerous private companies and public research institutions pursuing different pathways to fusion power. The success of these ventures is critical for meeting future energy demands and combating climate change, making the role of enabling technologies like those developed by Fusionality increasingly important. The company's focus on practical, deployable tools suggests a pragmatic approach to tackling the long-standing hurdles in the path of commercial fusion energy. While specific details regarding the proprietary nature of Fusionality's control systems and simulation platforms have not been fully disclosed, the company's stated objective is to democratize access to cutting-edge development tools for the fusion industry. This could potentially lower the barrier to entry for new fusion ventures and accelerate the overall pace of innovation. The founders' experience in AI and complex system modeling positions them to address some of the most computationally intensive and control-critical aspects of fusion reactor operation. The ultimate goal is to contribute to a future where fusion power plants are a significant component of the world's clean energy portfolio, providing a sustainable and virtually limitless energy source.
Financial Times · Sep 9, 2026
Anthropic withheld latest AI model from UK testing agency
Anthropic, a leading artificial intelligence company, has withheld its latest AI model, Claude 3.5, from safety testing by the UK's AI Safety Institute. This marks the first time Anthropic has opted out of such a testing program with the UK government. The decision has generated concerns within the British government that it could signal a broader trend of protectionism among major AI developers, potentially impacting the UK's ability to assess and regulate advanced AI systems. The AI Safety Institute, established in February 2024, is tasked with evaluating the risks posed by frontier AI models, including those capable of generating harmful content or exhibiting unpredictable behaviors. The institute's mandate includes conducting rigorous safety tests on new AI releases from prominent companies. Anthropic's refusal to participate means the UK will not have direct access to analyze Claude 3.5's capabilities and potential risks through its established governmental channels. This development comes at a critical juncture as governments worldwide grapple with the rapid advancement of AI and the need for robust safety frameworks. The UK government has been actively seeking to position itself as a global leader in AI safety, aiming to foster innovation while mitigating potential harms. The exclusion of Claude 3.5 from testing could hinder these efforts by limiting the institute's exposure to cutting-edge AI technologies. The AI Safety Institute was created with a budget of £100 million and has been actively engaging with AI developers to facilitate testing. Companies like OpenAI and Google DeepMind have previously cooperated with the institute. Anthropic's decision, therefore, stands out as a deviation from this collaborative approach. The implications of this move extend beyond the UK, as it could influence how other nations approach AI safety testing and regulatory cooperation with AI developers. The situation highlights the ongoing tension between the commercial interests of AI companies and the public interest in ensuring AI safety and security. The UK government has stated its commitment to understanding and addressing the risks associated with advanced AI, and the lack of access to Claude 3.5 for testing presents a significant challenge to this objective. The absence of testing data from Anthropic could leave a gap in the UK's understanding of the specific risks associated with Claude 3.5, potentially affecting its ability to develop effective policy responses. The AI Safety Institute's work is crucial for building public trust in AI and ensuring that the technology is developed and deployed responsibly. Anthropic's decision, while its specific reasons have not been fully detailed, raises questions about the future of international collaboration in AI safety governance.
TechCrunch · Sep 8, 2026
Cognition hits $48B valuation, signaling investors believe AI coding is far from a winner-take-all market
AI coding startup Cognition announced a valuation of $48 billion on May 15, 2024, following a new funding round. This significant valuation suggests that investors perceive the artificial intelligence coding market as one where multiple players can succeed, rather than a scenario dominated by a single winner. The company's valuation multiple is notably higher than that of Cursor, another AI coding tool, prior to its acquisition by SpaceX. Cognition's flagship product, Devin, is an AI software engineer designed to automate complex coding tasks, including debugging, writing code, and executing tests. Devin's capabilities are showcased through various benchmarks and demonstrations, where it has reportedly completed tasks that would typically require human engineers. For instance, Cognition has presented evidence of Devin successfully navigating and completing software engineering projects, such as finding and fixing bugs in open-source repositories. The company aims to position Devin as a comprehensive solution for software development, capable of handling the entire software development lifecycle. This includes understanding project requirements, planning development steps, writing code, testing, and debugging, thereby aiming to significantly enhance developer productivity and accelerate software delivery. The substantial investment in Cognition underscores a broader trend of increased investor interest in specialized AI applications. While large language models (LLMs) like those from OpenAI and Google have demonstrated general coding assistance capabilities, Cognition's focus on a dedicated AI software engineer highlights a market segment seeking more autonomous and sophisticated AI agents for complex technical tasks. The competitive landscape for AI coding tools includes established players and emerging startups, each vying to capture market share by offering unique features and performance advantages. Cognition's high valuation implies that investors believe Devin and similar specialized AI agents can carve out significant niches and achieve substantial market penetration. Cognition was founded by Scott Wu and other former Google DeepMind researchers. The company's success in attracting such a high valuation in a relatively short period reflects the rapid advancements and commercialization of AI technologies. The market for AI-powered development tools is expected to grow substantially, driven by the demand for faster development cycles, improved code quality, and the need to address software engineering talent shortages. Cognition's valuation suggests that the market is not yet saturated and that innovative solutions like Devin have considerable room for growth and impact within the software development industry.