Google has released the preview version of Gemini 3.1 Pro, optimized for complex reasoning and agent tool usage. It achieves 77.1% on ARC-AGI-2, 94.3% on GPQA Diamond, and 80.6% on SWE-Bench Verified. The model supports up to 1 million tokens input and 65k output, and provides a customtools preview endpoint. It is now available via Gemini API, AI Studio, and Vertex AI, and has entered public preview in GitHub Copilot. Pricing is per million tokens: $2 input / $12 output under 200K, and $4 input / $18 output beyond 200K.

TechCrunch reports OpenAI is nearing a financing round exceeding $100 billion, with a potential valuation above $850 billion, up from a prior expectation of around $830 billion; pre-money valuation for this round is approximately $730 billion. Major investors reportedly include Amazon (up to $50B), SoftBank (~$30B), NVIDIA (nearly $20B), and Microsoft. Funds will support rising training and product costs and provide cash runway toward profitability. The company is also testing ad monetization within ChatGPT.

India's Reliance Industries has announced a $110 billion investment over seven years to build AI computing infrastructure, including gigawatt-scale data centers and a nationwide edge computing network, integrating AI services with its Jio telecom ecosystem. Construction of multi-gigawatt data centers has begun in Jamnagar, Gujarat, with over 120 megawatts of capacity expected online by late 2026. The project will leverage approximately 10 gigawatts of surplus solar power from Gujarat and Andhra Pradesh. Reliance also plans to drive AI adoption in manufacturing, agriculture, healthcare, finance, and develop multilingual Indian language capabilities.

World Labs, co-founded by Fei-Fei Li, has announced a new $1 billion funding round, with investors including AMD, NVIDIA, Autodesk, and Fidelity. Reports indicate Autodesk contributed $200 million in a single investment. The company focuses on 'spatial intelligence' world models and has launched its product Marble, which generates and edits persistent, interactive high-fidelity 3D environments from text, images, video, or 3D layouts. It supports export in mesh, Gaussian point cloud, and video formats, offering both free and paid versions. Target markets include gaming, visual effects, VR, and robotics simulation—workflows requiring long-term editable world states.

OpenAI and Paradigm have launched EVMbench, a benchmark to evaluate AI agents' ability to discover, fix, and exploit vulnerabilities in smart contracts. GPT-5.3-Codex scored 72.2% on exploit tasks, a significant improvement from GPT-5's 31.9% about six months earlier. However, overall repair performance remains below 50%, though success rates in patching can increase from 39% to 94% with prompt assistance. OpenAI is also providing $10 million in API credits to support defensive auditing research and has introduced isolation tools like the Firecrawl Browser Sandbox to reduce risks from agent execution.

Cisco's 'State of AI Security 2026' report reveals that 83% of organizations plan to deploy AI agents, but only 29% believe they can use them securely, with governance and testing lagging and increasing attack surface. The report highlights risks including prompt injection, model jailbreaking, AI supply chain vulnerabilities, and risks associated with Model Context Protocol (MCP)-style agent systems, while comparing policy approaches across the U.S., Europe, and China. Cisco also released several open-source tools to detect vulnerabilities in open-weight models and signs of supply chain attacks, helping security teams conduct risk assessments and prioritize responses amid growing cloud adoption and agent deployment.

Google/DeepMind has integrated the music generation model Lyria 3 into the Gemini app, enabling users to generate ~30-second music clips from text or images, along with synchronized lyrics, music, and cover art. It supports eight languages and uses SynthID to watermark generated content. Free users can generate up to 10 tracks per day, with higher quotas for paid users. This integration embeds generative audio directly into a general-purpose AI interface, signaling a shift from standalone tools to platform-level distribution of music generation. It also establishes default mechanisms for content compliance and traceability.