What changed in 2025

Mistral released mistral-ocr-2512, its third OCR model, on 18 December. It turns scanned pages and PDFs into text a model can work with. If you feed documents to a model, it is one more reader to compare on your own files.

Google has officially launched Gemini 3 Flash, the newest entry in its generative AI model family. This version is specifically engineered for high-speed performance and low-latency responses, making it ideal for real-time applications and high-volume data processing. It maintains the core reasoning improvements found in the Gemini 3 architecture while significantly reducing operational costs for developers.

The model introduces an expanded 1-million-token context window, allowing for the processing of massive datasets, long documents, and extended video files in a single pass. Gemini 3 Flash also features enhanced multimodal capabilities, offering better native understanding of audio and visual inputs compared to its predecessors. It is now available for integration through Google AI Studio and Vertex AI.

Anthropic has announced the immediate availability of Claude Opus 4.5, its most advanced model designed for software engineering, autonomous agents, and computer use. The model sets new benchmarks on SWE-bench Verified and demonstrates significant improvements in handling ambiguity and multi-step reasoning tasks. Developers can access the model via the API using the identifier claude-opus-4-5-20251101, and it is also available across all major cloud platforms.

The new release features a revised pricing structure of 5 dollars per million input tokens and 25 dollars per million output tokens, making frontier-level reasoning more cost-effective. Alongside the model, Anthropic updated the Claude Developer Platform and Claude Code, introducing tools for long-running agents and enhanced integrations for Excel and desktop environments. Early testing indicates substantial efficiency gains, with some workflows requiring up to 65 percent fewer tokens while maintaining superior performance in code refactoring and long-horizon tasks.

Google has officially launched Antigravity, a comprehensive development platform designed for professional, frontend, and fullstack builders. The platform aims to streamline the development lifecycle by providing intuitive workspaces and integrated tools for various engineering needs. Recent updates have already expanded the platform's capabilities with the integration of Gemini 3 Flash and Nano Banana Pro models. These additions provide developers with advanced generative AI features and specialized model support directly within the Antigravity ecosystem.

Anthropic has released Claude Sonnet 4.5, its most powerful AI model to date, specifically optimized for complex coding and building autonomous agents. The model has achieved state-of-the-art results on the SWE-bench Verified evaluation and the OSWorld benchmark for computer use, where it reached a 61.4% success rate. It shows significant improvements in reasoning, math, and domain-specific knowledge across finance, law, and medicine compared to previous versions. Pricing for the new model remains consistent with Claude Sonnet 4 at $3 per million input tokens and $15 per million output tokens.

Alongside the model release, Anthropic introduced the Claude Agent SDK to provide developers with the infrastructure needed to build sophisticated agents. Several product updates were also announced, including a native VS Code extension for Claude Code and a new memory tool for the Claude API that allows agents to handle greater complexity. Within the Claude apps, users can now execute code and create files like spreadsheets and slides directly in the conversation. The Claude for Chrome extension has also been made available to Max users who were on the waitlist.

Thinking Machines Lab, founded by CTO Mira Murati, has secured a $2 billion Series A financing led by Andreessen Horowitz, with strategic participation from NVIDIA, Accel, ServiceNow, Cisco, AMD, Jane Street, and others aligned with its mission. The company’s goal is to accelerate collaborative general intelligence by building multimodal AI systems that seamlessly integrate conversation, vision, and real-world collaboration. Over the coming months, Thinking Machines Lab will unveil its first product—featuring a substantial open-source component—designed to equip researchers and startups with customizable AI building blocks. Additionally, it plans to release its core scientific breakthroughs to foster deeper understanding of frontier AI architectures. To support rapid development, the company is actively recruiting talented engineers and researchers committed to transforming cutting-edge research into practical, widely distributed tools that enhance individual agency.

Mistral AI has released Voxtral, an open-source suite of automatic speech recognition (ASR) models that deliver state-of-the-art accuracy and multilingual support as an alternative to both high-error open-source systems and expensive proprietary APIs. The initial offering comprises two variants—a 24 billion-parameter model optimized for production-scale deployments and a 3 billion-parameter mini model designed for edge and local use cases—both licensed under Apache 2.0. Voxtral models provide native multilingual transcription, extensive context processing, and robust error resilience, enabling seamless integration via Mistral’s API and full control over deployment and data privacy at less than half the cost of leading cloud ASR services. Available immediately through Mistral AI’s developer API and on GitHub, Voxtral sets a new benchmark for open speech recognition performance.

Anthropic has introduced Claude for Financial Services, a comprehensive AI-driven solution designed to transform financial analysis by unifying disparate datasets into a single, conversational interface. Leveraging the open-standard Model Context Protocol, Claude integrates trusted data from leading providers—including S&P Global’s Capital IQ Financials and earnings transcripts, Morningstar, FactSet, Databricks, Snowflake, Box, Daloopa, and Palantir—enabling analysts to perform market research, due diligence, and reporting with natural-language queries. The platform automates complex analytical workflows, reducing hours of manual data aggregation and cross-referencing to mere minutes, while ensuring every insight is verifiable through direct links to source materials. Built for enterprise-scale workloads, Claude for Financial Services offers expanded capacity and rigorous security protocols, guaranteeing that client data remains private and is not used to train underlying AI models. With its public launch on July 15, 2025, Anthropic aims to streamline investment decision-making and set a new standard for AI-powered financial services.

Cognition AI has signed a definitive agreement to acquire Windsurf, the developer-centric AI-powered Integrated Development Environment (IDE), in a move to bolster its enterprise software and AI-driven coding capabilities. The acquisition follows a dramatic 72-hour period during which OpenAI’s $3 billion offer for Windsurf expired and Google executed a $2.4 billion reverse-acquihire for Windsurf’s CEO and co-founders—highlighting the intense competition for top AI talent and technology in the developer tools space. Windsurf, valued at $1.25 billion and backed by investors including Kleiner Perkins and General Catalyst, achieved $82 million in annual recurring revenue with over 350 enterprise clients and hundreds of thousands of daily active users. Although financial terms of the deal were not disclosed, the acquisition encompasses Windsurf’s core IP, product line, trademark, brand, and the remaining engineering, product, and go-to-market teams. In the near term, Windsurf will continue to operate independently while Cognition invests heavily in integrating Windsurf’s IDE capabilities with its flagship autonomous coding agent, Devin. This combined offering aims to enable developers to plan tasks, delegate code generation to AI agents, and review pull requests seamlessly within a unified interface. Longer-term integration roadmaps will bring deeper spec-driven workflows and enhanced developer experiences across both platforms.

Google has made its first Gemini Embedding text model (gemini-embedding-001) generally available to developers via the Gemini API and Vertex AI. First introduced experimentally in March 2025, this embedding model has consistently ranked at the top of the Massive Text Embedding Benchmark (MTEB) Multilingual leaderboard, outperforming both Google’s legacy embeddings and other proprietary offerings across tasks such as semantic retrieval and classification in domains like science, legal, finance, and coding. The model supports over 100 languages with a maximum input length of 2,048 tokens and leverages Matryoshka Representation Learning (MRL) to allow teams to choose output dimensions of 3,072, 1,536, or 768 vectors—optimizing for accuracy, compute, and storage trade-offs. Developers can begin experimenting at no cost on the free tier, with production usage priced at $0.15 per 1 M input tokens; the experimental gemini-embedding-exp-03-07 will be supported until August 14, 2025 before deprecation. Integration is seamless using the existing embed_content endpoint in Google AI Studio, and batch processing support for asynchronous embedding workflows is coming soon.

Amazon has unveiled Kiro, an AI-powered integrated development environment designed to guide developers from initial prompt to production-ready applications with minimal friction. By leveraging autonomous AI agents, Kiro translates a single natural-language instruction into comprehensive specification artifacts—user stories, technical design docs, and task sequences—ensuring every assumption is explicit and traceable. Its two core capabilities, Kiro specs and Kiro hooks, automate upfront planning and background tasks such as test generation, documentation updates, and security validations, catching issues before code is committed. Built on the open-source Code OSS foundation, Kiro preserves familiar VS Code settings and plugin compatibility while offering a unified “vibe coding” experience backed by rigorous spec-driven workflows. Released in public preview on July 14, 2025, Kiro aims to bridge the gap between rapid AI-generated prototypes and robust, maintainable production systems.

Moonshot AI has released Kimi K2, a groundbreaking open-source large language model built on a Mixture-of-Experts (MoE) architecture with one trillion total parameters and 32 billion activated parameters. Trained on over 15.5 trillion tokens, Kimi K2 excels at complex task decomposition, native tool integration, coding, and frontier knowledge reasoning—outperforming leading open-source peers like DeepSeek’s V3 and rivaling proprietary U.S. models in key benchmarks. Two variants are available: Kimi-K2-Base for researchers and developers seeking full fine-tuning control, and Kimi-K2-Instruct for drop-in, general-purpose chat and agentic workflows. Released under a permissive MIT-style license, the model is freely accessible via web and mobile applications, inviting the global developer community to innovate, extend, and deploy advanced AI capabilities without barriers.

Google DeepMind (via Google) introduced GenAI Processors, an open‑source Python library released on July 10, 2025. This toolkit offers a unified Processor interface for handling asynchronous, stream‑based AI workflows—spanning input pre‑processing, model calls, and post‑processing. It supports concurrent execution of multimodal streams (e.g., audio, video, text), enabling low-latency pipelines like live AI agents with the Gemini API. The library is available under Apache 2.0 on GitHub.

Perplexity has unveiled Comet, an AI-powered web browser built on Chromium and offered to its Perplexity Max subscribers. With a hefty $200/month introductory price, Comet transforms browsing into an interactive, conversational experience—integrating its AI sidebar for tasks like booking meetings, sending emails, or shopping. The browser also features first-party support for natural-language querying of browsing history and cross-session memory. Comet is available now on Windows and Mac, with broader access planned soon via an invite system.

xAI, Elon Musk’s AI firm, officially launched Grok 4 (and a premium “Heavy” variant) on July 9 2025, positioning it as a major leap in reasoning, multimodal capabilities, and coding assistance. Built on the Colossus supercomputer, Grok 4 boasts PhD‑level performance on academic benchmarks, a massive 256k‑token context window, real‑time web search via X integration, and a specialized Grok 4 Code model for developers. The release follows hot on the heels of controversies around hate‑speech outputs from previous versions—xAI emphasized updated content‑moderation safeguards. Pricing starts at $30/mo, with “Heavy” at $300/mo.

Cursor issued a formal apology following widespread criticism over its new pricing changes and drastic rate-limit reductions. The company admitted that the rollout was poorly communicated and caused confusion and unexpected charges for many users between June 16 and July 4. Cursor has offered refunds to affected customers (billed extra) and acknowledged that trust was damaged by how the changes were handled.

Despite the apology and promise of refunds, it remains unclear whether Cursor will restore previous request quotas or make further adjustments to rate limits. Many developers are still frustrated, pointing out that the $20 credit now only covers about 225 Sonnet 4 requests, compared to 500 previously promised. The developer community continues to express dissatisfaction, citing persistent rate-limit errors, lack of transparency, and inadequate communication.

On Friday, July 4, 2025, Elon Musk posted on X (formerly Twitter) stating, "We have improved @Grok significantly. You should notice a difference when you ask Grok questions." This update is part of xAI's ongoing efforts to enhance Grok's capabilities, particularly in challenging what Musk perceives as ideological biases in other AI models. While specific technical details of this immediate improvement were not fully disclosed, it follows recent discussions and reports about the upcoming Grok 4, which is expected to launch shortly after July 4th and will feature advanced reasoning, specialized coding capabilities, and a re-training on a "cleaner" knowledge base.

Channel 4 became the first UK broadcaster to air an AI-generated advertisement on its streaming platform. The campaign was created with an in-house generative AI tool that combines scripted prompts, AI video synthesis, and automated voiceover. Human teams provided creative direction and conducted compliance checks to ensure brand safety and regulatory adherence. The offering is part of Channel 4’s new service aimed at helping small and medium-sized businesses produce professional ads quickly and affordably without large creative budgets.

The Certainty Project report published on June 30, 2025, revealed that 18 percent of mid-sized companies are now integrating real-time analytics and scenario modeling into their core strategy planning. These firms are using machine learning simulations to forecast the impact of trade policy shifts, supply chain disruptions, and macroeconomic volatility. This trend shows a clear shift from periodic reporting to continuous decision support, as executives seek to make proactive adjustments in rapidly changing markets. The findings highlight how mid-market organizations are investing in data-driven capabilities once reserved for large enterprises.

At the Open Source Summit North America in June, Google donated its Agent2Agent (A2A) protocol to the Linux Foundation. Designed to standardize communication between AI agents, the protocol fosters interoperability across tools and platforms. Building on ecosystem efforts like Anthropic’s Model Context Protocol, A2A marks a key step toward an open, collaborative AI agent ecosystem across open-source and enterprise solutions.

Amazon announced a $10 billion investment in expanding US data center infrastructure in North Carolina. The move aims to support growing AI and cloud workloads, create hundreds of local jobs, and strengthen AWS’s position as a leader in high-performance computing services.

Google officially launched Gemini CLI, an open-source AI agent that brings the power of Gemini directly into developers' terminals. It provides lightweight access to Gemini, allowing developers to use natural language for a wide range of tasks, including coding, debugging, content generation, and deep research, with "unmatched usage limits" for individual developers. It leverages the Gemini 2.5 Pro model with its massive 1 million token context window and integrates with Google's AI coding assistant, Gemini Code Assist.

A U.S. District Court for the Northern District of California issued a landmark ruling that training generative AI models on copyrighted books qualifies as fair use under U.S. copyright law. The court described model training as “exceedingly transformative” because it produces new, non-infringing outputs. However, it excluded the question of whether storing and distributing unauthorized copies of books in training datasets was legal, leaving that issue for future litigation. The decision was widely seen as a significant win for companies like Anthropic and OpenAI, while authors’ groups expressed disappointment over the implications for creator compensation.

Pure Storage used its annual Pure Accelerate conference in Las Vegas to unveil Enterprise Data Cloud, a new platform designed to centralize data management, governance, and compliance across hybrid and multi-cloud environments. The solution features Fusion, an intelligent control plane that automates provisioning, scaling, and optimization of data storage resources. Enterprise Data Cloud is built with AI Copilot assistance to streamline complex tasks like workload balancing and policy enforcement. The launch also included enhancements to FlashArray systems to improve performance and reliability for AI training workloads.

Google announced the general availability of Gemini 2.5 Flash and Gemini 2.5 Pro, offering developers a range of models optimized for speed, cost, and performance. Gemini 2.5 Pro supports a 1 million-token context window, advanced code understanding, and retrieval-augmented generation, while Flash is tuned for fast, low-latency tasks. A preview of Flash‑Lite—a smaller, even more efficient variant—was also introduced as part of Google Cloud’s Gemini API.

Adobe announced the LLM Optimizer, a new Experience Cloud feature introduced at the Cannes Lions International Festival of Creativity. The tool enables marketers to track how their content appears across AI-powered experiences, including chatbots, voice assistants, and search engines. Brands can analyze AI-generated summaries, identify performance gaps, and adjust messaging strategies to improve visibility. Adobe also upgraded GenStudio with enhanced generative AI capabilities for creating videos, display ads, and social content tailored to AI-centric discovery platforms.

At the Gartner Data & Analytics Summit in Mumbai, industry leaders tackled critical issues like recruiting top talent, clearing up AI misconceptions, and navigating fragmented data ecosystems. Sessions emphasized the growing importance of intelligent decision-making frameworks and reinforced the value of data-driven strategies in today’s business landscape.

Krutrim has officially introduced Kruti, India’s first multilingual agentic AI assistant built to handle everyday tasks in 13 Indian languages. Kruti blends advanced conversational intelligence with contextual memory and API integrations, empowering users to book rides, order groceries, pay bills, and manage daily activities through simple voice or text commands. This launch marks a significant milestone in creating culturally tailored AI experiences for millions of Indian users.

Databricks announced the public preview of Lakebase, its fully managed PostgreSQL engine embedded directly into the Lakehouse platform. Lakebase is designed for sub-10 millisecond latency and up to 10,000 queries per second, providing fast transactional workloads alongside data lake analytics. It also supports Git-style branching to enable safe experimentation and reproducible pipelines, making it easier for teams to manage complex AI data workflows. The service integrates with Unity Catalog for governance, offering unified data permissions across lakehouse and relational assets.

OpenAI has launched o3‑pro, an enhanced version of its flagship o3 reasoning model, now available to ChatGPT Pro and Team users and in the API. Designed for reliability over speed, o3‑pro excels in precision-heavy tasks, including advanced coding, math, science, and thoughtful instruction-following. It features expanded tool support—web search, file analysis, visual reasoning, Python execution, and personalized memory—though response times are longer than lighter models. In OpenAI's internal “4/4 reliability” benchmarks, o3‑pro outperformed o1‑pro and base o3. Initially, certain features like temporary chats, image generation, and Canvas are disabled. Model pricing via API is $20 per million input tokens and $80 per million output tokens.

At the Data + AI Summit in San Francisco, Databricks unveiled "Agent Bricks" (beta)—modular templates to quickly build AI agents—along with LakeFlow Designer to simplify ETL and ML pipelines, and MLflow 3.0 to boost model tracking at scale. They also enhanced security, compliance, and governance integrations.

Apple introduced Live Translation, enabling real-time voice and text translation in FaceTime, Messages, and Phone, powered entirely by on-device AI. This feature maintains user privacy by handling translations locally and offers seamless multilingual communication.

Apple introduced Liquid Glass, a translucent and dynamic interface design spanning iOS 26, iPadOS 26, macOS Tahoe, watchOS 26, tvOS 26, and visionOS 26. This fresh visual language brings glass-like reflections and responsive layering to widgets, controls, and icons—drawing design inspiration from visionOS and the Vision Pro headset.

Databricks revealed it had acquired Neon, a fast-growing serverless PostgreSQL company, for $1 billion. Neon’s technology allows developers to create instantly branching, scalable databases with Git-like workflows, making it easier to spin up test and production environments for AI applications. Integrating Neon into Databricks’ Lakehouse platform is intended to help customers build and deploy AI-powered analytics solutions more quickly while reducing operational overhead. The acquisition underscores Databricks’ push to blend traditional databases with modern data lake architectures to serve machine learning use cases end to end.

Snowflake announced it had acquired Crunchy Data, a leader in enterprise PostgreSQL solutions, for $250 million. The acquisition expands Snowflake’s capabilities in open-source relational databases and strengthens its position as a unified platform for both structured and unstructured data. With Crunchy Data’s expertise in secure, production-grade PostgreSQL, Snowflake aims to deliver more robust AI and machine learning workloads while appealing to developers who rely on open standards. This move reflects a broader trend of cloud data companies consolidating to meet growing demand for AI-ready infrastructure.

OpenAI's Codex, a cloud-based AI software engineering agent, is now available to ChatGPT Plus users. This availability for Plus users signifies a broader push by OpenAI to make advanced AI coding assistance more accessible.

Apache Spark 4.0.0, the latest major release of the popular unified analytics engine, was officially released on May 23, 2025. This release marks a significant milestone in the evolution of Spark, bringing substantial advancements across the board to enhance performance, accessibility, and developer productivity for large-scale data analytics and AI workloads.

Anthropic has officially released Claude Code, a powerful, coding-focused extension of the Claude model family. Originally previewed in February with Claude 3.7 Sonnet, this agentic tool is now generally available and integrates seamlessly into developer workflows. It supports background tasks via GitHub Actions and provides native extensions for VS Code and JetBrains. Claude Code enables inline file edits, automated debugging, and direct collaboration inside your IDE, empowering developers to delegate substantial engineering tasks to AI while maintaining full control and visibility.

Anthropic has released Claude 4, the next-generation AI model family featuring Claude Opus 4 and Claude Sonnet 4. This major upgrade brings hybrid reasoning—with both instant replies and deep “extended thinking”—as well as parallel tool usage, improved memory, enhanced coding performance, and agentic capabilities that allow models to maintain focus for hours. Opus 4 sets new benchmarks in coding while Sonnet 4 offers a cost-effective balance of power and speed.

OpenAI's Codex is an artificial intelligence model designed to translate natural language into code. Essentially, you can tell it what you want to achieve in plain English, and it will generate programming code to accomplish that task.

GitHub Copilot is rolling out Agent Mode with MCP support to all VS Code users. This update enhances Copilot's capabilities, allowing it to take action on user prompts by completing necessary subtasks and suggesting terminal commands. The Model Context Protocol (MCP) enables Agent Mode to access various tools and context, providing more interactive coding support. Additionally, GitHub has announced a new Copilot Pro+ plan and the general availability of models from Anthropic, Google, and OpenAI. Users can also now access next edit suggestions and the Copilot code review agent.

nowflake announced the general availability of Cortex COMPLETE Structured Outputs. This feature allows users to define desired outputs and their formats via a schema, simplifying prompting, reducing the need for post-processing in AI data pipelines, and enabling seamless integration with systems requiring deterministic responses.

Global food and beverage company Danone adopted the Databricks Data Intelligence Platform to serve as its core data foundation for analytics and AI. This integration is expected to improve data accuracy and reduce 'data-to-decision' time by up to 30%, enabling smarter and faster decision-making across the organization.

Financial services firm Macquarie began coverage of Snowflake, assigning a "Neutral" rating. The assessment reflects a balanced view of Snowflake's growth prospects in the competitive cloud data platform market.

Databricks announced the general availability of Lakeflow Connect, a no-code data ingestion tool that enables seamless integration from SaaS applications and databases into the Databricks platform. This enhancement aims to simplify data workflows for users.

xAI has acquired X in an all-stock deal valuing the combined entity at $113B, bringing together xAI’s cutting-edge AI models and infrastructure with X’s massive 600M+ user base and real-time content ecosystem. The merger integrates data, distribution, compute power, and top-tier talent to build a platform that not only reflects the world but actively drives innovation and accelerates human progress.

Alibaba introduced Qwen2, an open-source AI model designed to power cost-effective AI agents. Qwen2 supports multiple languages and can run on low-resource environments, making it ideal for startups and developers building scalable AI tools.

Databricks and Anthropic have announced a five-year partnership. This will bring Anthropic's Claude models directly to the Databricks Data Intelligence Platform. This integration will allow over 10,000 customers to build and deploy AI agents that can understand and process their own business data. The latest Claude 3.7 Sonnet model is now available on Databricks across major cloud platforms. The partnership aims to help businesses get more value from their AI investments by making it easier to create, use, and manage AI agents securely. Key advantages include the ability to build specialised AI agents, seamless integration of Claude models into the Databricks platform, and unified governance for data and AI development. This collaboration seeks to unlock the full potential of enterprise data through AI.

Gemini 2.5 Pro, which Google DeepMind considers their most intelligent AI model. It improves upon the existing Gemini models by including native multimodality and a long context window. A key feature is the models' ability to reason through their thoughts before generating a response, which leads to better performance and accuracy. The Gemini 2.5 Pro Experimental version is specifically highlighted as being excellent for coding and handling complex prompts, and is currently their most advanced coding model. This model achieves significantly improved results across various benchmarks, reaching state-of-the-art in maths and science evaluations. It also offers a 1-million token context window. Developers can access and experiment with Gemini 2.5 Pro Experimental via Google AI Studio and the Gemini API. The model supports a wide range of inputs, including text, images, video, and audio, with text being the output format.

OpenAI unveiled GPT-4o's native image generation feature, marking a significant advancement in AI-driven content creation. This integration allows GPT-4o to produce precise, accurate, and photorealistic images directly within ChatGPT, enhancing the model's multimodal capabilities. Users can now generate and edit images seamlessly through natural language prompts, facilitating applications in design, education, and creative storytelling. This development underscores OpenAI's commitment to creating versatile AI tools that cater to a broad spectrum of user needs.

CEO Jensen Huang revealed upcoming advancements in AI hardware, including the Blackwell Ultra graphics architecture and Vera Rubin AI chip. He projected significant growth in data center revenue driven by demand for GPUs.

Apple announced on March 7, 2025, that the rollout of significant AI enhancements to its voice assistant, Siri, has been delayed until 2026. The company stated that it is working on a more personalized version of Siri that would better understand user context and perform tasks across various applications. However, internal testing revealed that the features were not functioning reliably enough for release, prompting the delay.

OpenAI has just released GPT-4.5, marking a significant leap forward in conversational AI. This new model not only excels in creative tasks and everyday interactions but also brings a fresh edge by offering blunt, even edgy responses when requested. According to Sam Altman, GPT-4.5 is the first model that truly feels like conversing with a highly attentive person. With improvements that make it three times better at fact-checking and far less error-prone than GPT-4, and enhanced capabilities in understanding context, tone, and emotion, GPT-4.5 is set to change the game in how we interact with AI.

Amazon has unveiled its latest advancement in voice technology with the release of Alexa+. This innovative update leverages cutting-edge generative AI to enhance Alexa's natural language understanding, enabling more dynamic and context-aware interactions. With this upgrade, Alexa delivers richer conversational experiences, improved response accuracy, and a more intuitive user interface.

Anthropic has unveiled Claude 3.7 Sonnet, its latest breakthrough in conversational AI. This advanced model features hybrid reasoning, allowing it to deliver both rapid responses and detailed, step-by-step analysis in a single interaction. With enhanced coding accuracy—achieving 70.3% on the SWE-bench Verified benchmark—and the ability to generate outputs up to 128,000 tokens, Claude 3.7 Sonnet is poised to revolutionize content creation and software development. Additionally, Anthropic introduces Claude Code, an innovative coding assistant designed to collaborate with developers by searching, editing, and managing code directly from your terminal. Claude 3.7 Sonnet is now accessible via the Claude app, Anthropic’s API, Amazon Bedrock, and Google’s Vertex AI.

Satya Nadella announced a quantum computing breakthrough, creating a new state of matter using topoconductors to power the Majorana 1 quantum processing unit. This advancement allows for faster, more reliable, and smaller qubits, about 1/100th of a millimetre, paving the way for a million-qubit processor. This technology aims to solve problems beyond the capabilities of current computers, with the goal of creating a meaningful quantum computer in years, not decades. The focus is on building technology that increases productivity and benefits economies globally.

Databricks now offers automatic liquid clustering for Unity Catalog managed tables, available on DBR 15.4 and above. This innovative feature intelligently selects optimal clustering keys by continuously monitoring query patterns and data distribution, dynamically optimizing the data layout to improve query performance. By adapting to evolving workloads, automatic liquid clustering reduces latency and minimizes the need for manual tuning. With this public preview, data engineers can enjoy enhanced efficiency and lower operational overhead while ensuring that their tables are always optimally organized for rapid data access.

Grok is an AI tool that offers unfiltered answers and advanced capabilities in reasoning, coding, and visual processing. It stands out with its ability to generate images from text, providing photorealistic rendering. Grok enhances understanding with real-time insights and can decipher documents and images, converting visuals into insights. Additionally, it assists with productivity tasks like essay writing, coding, and problem-solving, and is available across web, iOS, and Android platforms.

OpenAI has significantly upgraded ChatGPT by enabling file and image uploads for both the o1 and o3-mini models. This enhancement allows users to interact with ChatGPT in richer ways, providing visual and document-based context for their conversations.

In addition to this, OpenAI has also dramatically increased the usage limits for ChatGPT Plus users on the o3-mini-high model. The previous limit has been multiplied by seven, now allowing Plus subscribers up to 50 uses per day. This substantial increase provides more flexibility and access to the advanced capabilities of the o3-mini-high model.

Databricks has announced that Databricks Runtime 16.2 and Databricks Runtime 16.2 ML are now generally available. This new release delivers significant improvements in performance, stability, and security, along with enhanced support for machine learning workloads. With these updates, users can optimize their data processing and ML model development workflows for faster and more efficient operations. For more details, see Databricks Runtime 16.2 and Databricks Runtime 16.2 for Machine Learning.

Deep research is a new agentic capability in ChatGPT designed for complex, multi-step research tasks, completing in minutes what would take humans hours. It independently finds, analyses, and synthesises online sources, leveraging reasoning to interpret diverse data like text, images, and PDFs. This tool generates comprehensive, well-documented reports with citations. Unlike GPT-4o, which is geared toward real-time conversations, deep research focuses on in-depth analysis and verified answers. Trained using reinforcement learning on difficult browsing and reasoning tasks, it can use tools such as a browser and Python to conduct research. It has achieved high scores on expert-level tests, demonstrating its ability to seek out specialized information. Although powerful, deep research has some limitations, such as the potential to hallucinate facts, struggle with differentiating authoritative sources, and make errors in formatting. It is currently available to Pro users with up to 100 queries per month, with access rolling out to Plus and Team users next.

OpenAI's o3-mini redefines AI by integrating advanced reasoning, structured thought processes, and safety-focused mechanisms. Unlike traditional models, it evaluates multiple approaches before formulating responses, ensuring accuracy and reliability. With a record-breaking performance in coding and problem-solving, o3-mini stands out as a powerhouse of intelligence. Its resistance to jailbreaks, reduced hallucination rates, and fairness evaluations make it one of the safest and most ethical AI models. Moreover, its multilingual proficiency enhances accessibility worldwide. As AI evolves, o3-mini sets a new standard for responsible, adaptive, and high-performing artificial intelligence.

In January 2025, Databricks released a number of new features and platform improvements. Some of the key updates include:

• Predictive optimization can now be enabled at the catalog or schema level.
• Filtering full datasets for large tables is now supported.
• Clean Rooms is now generally available, with new APIs for automation, self-collaboration within a single metastore, and support for output tables on Azure. Also, you can now create a Clean Room with a HIPAA compliance security profile.
• Delta Live Tables now supports publishing to tables in multiple schemas and catalogs.
• Databricks Runtime 16.2 is now available as a Beta release.
• OAuth token federation is now available in public preview.
• Notebook load times have been improved.
• Notebooks are now supported as workspace files.
• Failed tasks in continuous jobs are now automatically retried.

Apache Spark 4.0.0 is upcoming and a preview-2 version is available. Spark Connect is a major feature in this release. The preview version allows users to explore new features and debug from their local machines using hashtag#SparkConnect and hashtag#vsCode. The vision of running Apache Spark from anywhere is a key aspect of this release. The full release of Spark 4.0 will include enhanced usability and debuggability with Spark Connect's GA (Generally Available) release

Gemini's new 2.0 Flash model is now available for all users on web and mobile. This upgrade is designed for better agent-like interactions, providing faster responses and improved performance on common tasks like brainstorming, learning, and writing. Gemini Advanced users retain their existing benefits, including the large context window for file uploads, priority access to advanced features, extra storage, and more.

Gemini Advanced users will continue to have access to a 1M token context window for up to 1,500 pages of file uploads, priority access to features such as Deep Research and Gems, 2TB of storage, and more.

The future of Artificial Intelligence is no longer a distant dream—it’s being built right now with a scale and ambition like never before. The Stargate Project, a collaboration between tech giants like Oracle, SoftBank, and OpenAI, marks a monumental step toward revolutionizing AI and pushing the boundaries of what’s possible.

This isn’t just about technological progress; it’s about creating a foundation to advance Artificial General Intelligence (AGI)—AI systems capable of performing any intellectual task a human can. Stargate is not only focused on AI for today but is paving the way for AGI to benefit humanity as a whole.

DeepSeek isn’t just another player in the AI revolution—it’s a trailblazer pushing the boundaries of AI capabilities, accessibility, and reasoning. Think of it as a team of expert puzzle-solvers who are not only mastering AI development but also sharing their insights with the world. With groundbreaking models and innovative training techniques, DeepSeek is making AI smarter, more efficient, and widely accessible.