According to God of Prompt on Twitter, citing a Google AI experiment, standard fine-tuning of Gemini 2.5 Flash on a text-to-SQL dataset reduced performance on the hardest queries, indicating reasoning degradation without explicit reasoning traces. As reported by the tweet, the base Gemini 2.5 Flash scored 73.17% overall vs 72.50% after fine-tuning, but on the hardest 40 queries it fell from 62.5% to 57.5%, a failure mode Google calls representation collapse. According to the same source, a Qwen 7B model improved from 36.17% baseline to 45.33% with standard fine-tuning, and to 54.5% when trained with Chain of Thought steps, nearly halving the gap with Gemini 2.5 Flash. The business takeaway, according to the thread, is that large models risk losing multi-step reasoning when fine-tuned on plain IO pairs, while small models gain materially when trained on structured reasoning traces, making CoT-style fine-tuning and data format design a high-ROI strategy for enterprise text-to-SQL and analytics automation.

According to God of Prompt on X, Zhejiang University, Alibaba, Tencent, and 15 partner institutions introduced SkillNet, a shared library of 200,000+ reusable skills that any AI agent can call to avoid relearning each session. As reported by the X post, tests on DeepSeek V3, Gemini 2.5 Pro, and o4 mini across three environments showed an average 40% reward improvement and 30% fewer execution steps versus baselines, with immediate skill transfer requiring no retraining or parameter updates. According to the post, the repository includes 150,000+ curated skills evaluated on safety, completeness, executability, maintainability, and cost. If verified in broader benchmarks, this infrastructure could cut agent operating costs, shorten development cycles for autonomous workflows, and enable cross-model capability sharing for enterprise automation.

According to Jeff Dean on X, Gemini 3.1 Flash Lite is significantly faster in tokens per second than the older Gemini 2.5 Flash and completes complex tasks with roughly one third the tokens used in the comparison shown. As reported by Jeff Dean, the side-by-side demo indicates higher accuracy alongside speed and token savings, implying lower latency and reduced inference cost for production workloads. According to Jeff Dean, the reduced token usage can cut API spend and improve mobile and edge deployment efficiency where context windows and bandwidth are constrained. As reported by Jeff Dean, these gains suggest opportunities for upgrading chatbots, agents, and RAG pipelines to achieve faster response times, better user experience, and higher request throughput on existing infrastructure.

According to Jeff Dean on X, Gemini 3.1 Flash Lite delivers significantly higher token throughput and uses roughly one third the tokens to complete the same complex task compared with Gemini 2.5 Flash, based on his posted side-by-side speed and accuracy video comparison. As reported by Jeff Dean, the new model’s faster tokens-per-second and lower token usage indicate reduced inference latency and cost per task for production workloads, enabling cheaper summarization, agent loops, and multimodal reasoning at scale. According to the source video by Jeff Dean, the accuracy holds while token consumption drops, suggesting improved planning and compression that can cut prompt and output spend for enterprises deploying high-volume chat, RAG, and automation pipelines.

According to Sundar Pichai, the new Gemini 2.5 Computer Use model is now available in the Gemini API and has established a new standard across multiple AI benchmarks with improved low latency. The model’s standout feature is its advanced ability to interact with web elements such as scrolling, filling forms, and navigating dropdown menus, signaling a significant step toward developing general-purpose AI agents. Developers can access and test these advanced capabilities via API on Google AI Studio and Vertex AI, opening new business opportunities for automation and productivity tools (Source: Sundar Pichai on Twitter, Oct 7, 2025).

According to Google DeepMind, the new Gemini 2.5 Computer Use model leverages advanced visual understanding and reasoning to enable AI agents to navigate browsers by clicking, scrolling, and typing as a human user would. This upgrade significantly enhances practical AI applications for automated online tasks, streamlining workflows in industries such as customer support, e-commerce, and data entry. The model outperforms previous versions on multiple industry benchmarks, offering improved speed and reliability, which positions it as a game-changer for businesses seeking to automate complex web-based operations (source: Google DeepMind, Twitter, Oct 7, 2025).

According to Demis Hassabis, AI Mode for Search has officially launched in the UK, offering users enhanced search experiences through advanced reasoning, logical thinking, and multimodal understanding powered by Gemini 2.5 (source: @demishassabis). This update builds on previous AI Overviews, providing practical applications for both consumers and businesses, such as improved information retrieval, context-aware responses, and the ability to process multiple types of content including text and images. For AI industry players, the rollout signifies a major step in mainstreaming multimodal AI-powered search, opening up new opportunities for search engine optimization, targeted advertising, and integrating AI-driven customer interaction solutions.

According to Demis Hassabis on Twitter, Google has officially launched the Gemini 2.5 series, representing a significant advancement in AI model performance and usability. The new Gemini 2.5 models have been developed with direct user and developer feedback, resulting in improved accuracy, speed, and integration capabilities for enterprise and application use cases (source: Demis Hassabis, Twitter, June 17, 2025). This release opens new business opportunities for companies seeking scalable AI solutions for automation, data analysis, and generative AI applications. Enterprises and developers can now leverage Gemini 2.5’s enhanced features to build more sophisticated AI products and integrate cutting-edge AI into their workflows, accelerating innovation in sectors such as finance, healthcare, and e-commerce.

According to Demis Hassabis on Twitter, the unprecedented demand for Google's Gemini 2.5 and Veo 3 AI models is being successfully managed by their advanced infrastructure, chip, and Site Reliability Engineering (SRE) teams. This highlights the scalability and robustness of Google's custom Tensor Processing Units (TPUs), which are critical for supporting large-scale AI workloads. The operational excellence of these teams ensures continuous uptime and performance, enabling businesses to deploy state-of-the-art generative AI solutions at scale. As AI adoption accelerates, the ability to maintain and scale AI infrastructure such as TPUs presents significant business opportunities for cloud service providers and enterprise AI deployments (source: @demishassabis, Twitter, May 31, 2025).

According to Oriol Vinyals (@OriolVinyalsML), Gemini 2.5 has achieved the top position on a new AI benchmark leaderboard, highlighting its advanced performance in natural language processing tasks. This result, shared on Twitter on May 29, 2025, demonstrates Google's ongoing competitiveness in large language model development. For enterprises, Gemini 2.5's leadership on such benchmarks signals improved reliability and performance for AI-powered applications, potentially driving adoption in sectors like customer service automation, content creation, and enterprise data analysis. The benchmark achievement reinforces the need for businesses to continuously evaluate emerging AI models for integration opportunities in their workflows (source: Oriol Vinyals, Twitter).