Google DeepMind launches Genie 3, a real-time 720p world model enabling interactive virtual environments from text prompts, supporting multi-minute coherent navigation and scene modifications. This breakthrough technology for embodied AI training demonstrates dynamic simulations for robotics and open-ended tasks.

OpenAI partners with AWS to launch open-weight models on Bedrock and SageMaker, challenging Microsoft amid Oracle's 3B deal and AWS's shift from Anthropic.

Google launches AI agent suite to automate data workflows, including BigQuery tools for pipeline building, analytics, and multimodal data processing, with APIs and no extra charges.

OpenAI releases gpt-oss, its first open-source language models since GPT-2, featuring 117B/21B parameter variants under Apache 2.0. The MoE architecture reduces computational costs by 85%, with 120B matching o4-mini performance. Optimized for on-device and cloud deployment, the models offer commercial flexibility across regulated sectors while strategically countering competition from Chinese alternatives like DeepSeek-R1.

Singapore startup Sapient Intelligence introduces HRM, a 27M-parameter AI model outperforming billion-parameter systems in complex reasoning tasks using just 1,000 samples. The architecture employs hierarchical convergence and latent space computation to avoid chain-of-thought limitations, achieving 40.3% ARC-AGI accuracy versus 21.2% for Claude 3.

Elon Musk's xAI launches Grok Imagine, an iOS AI tool offering 15-sec Spicy Mode videos with explicit content capabilities. Despite adult prompt filters in testing, semi-nude imagery succeeds. Competing with Google DeepMind/OpenAI, the system faces rendering limitations and celebrity content restrictions while improving daily per Musk's updates.

Leading AI models including Gemini Pro 2.5, Claude Opus 4, and Grok 4 will compete in Google/Kaggle's three-day chess tournament starting August 5. The elimination format prohibits third-party tools, requiring 60-minute move decisions. Kaggle's game arena will benchmark LLM strategic capabilities through live-streamed matches and unscheduled simulations, with rankings based on both tournament results and hundreds of behind-the-scenes games.

ChatGPT nears 700M weekly active users, driven by GPT-4 visual module and enterprise adoption. Paid users exceed 5 million while daily engagement reaches 16 minutes, per Sensor Tower data.

OpenAI introduces Learning Mode in ChatGPT to enhance education via interactive problem-solving. The mode uses guided prompts, structured responses, and skill-level adaptation developed with educators. Features include Socratic questioning, modular content, personalized feedback, and seamless mode switching. Available to logged-in users, it aims to foster critical thinking over direct answers.

Apple is building an AI search engine via its AKI team to replace Google, targeting integration with Siri and Spotlight. Led by Robbie Walker, the project aims to reduce Google's $2B annual payment and ad dependency, potentially shifting to subscription models. While lagging behind Google and Microsoft, Apple seeks to leverage privacy and direct answers. Recent hires in search algorithms highlight the initiative's scale amid internal debates.

xAI launches Grok Imagine, enabling text-to-video generation up to 6 minutes via X platform, using Aurora model. Prioritizing SuperGrok subscribers through waitlist, while 'Spicy Mode' sparks controversy over explicit content risks.

Apple CEO Tim Cook emphasized AI urgency in an internal meeting, acknowledging Siri delays and competitive lags while asserting Apple's capacity to innovate modern tech, citing past breakthroughs.

Anthropic blocks OpenAI's full access to Claude models over policy violations but permits limited benchmarking access. The companies dispute usage terms, with Anthropic citing competitor restrictions and OpenAI calling the access "industry standard."

Apple is building an AI-driven "answer engine" team to develop a ChatGPT-like app, codenamed "Answers, Knowledge and Information," potentially integrating it into Siri/Safari. The project involves hiring search algorithm experts, as existing AI upgrades for Siri face delays and antitrust-related partnership renegotiations with Google.

Google DeepMind's Gemini 2.5 Deep Think, a multi-agent AI system, won IMO gold with enhanced reasoning capabilities. Available via $250/month subscription, it outperforms rival models in math benchmarks and code generation while requiring higher computational costs. The company is expanding access to researchers and enterprises through its API.

Manus launches Wide Research, its major update since 2024, with a $199/month tier. The feature enables collaborative AI agents to handle large-scale research and complex tasks like stock analysis, outperforming rivals like OpenAI. Co-founder to demo via video.

Meta's Q2 results beat forecasts with $7.14 EPS and $47.5B revenue, driven by $46.56B ad revenue growth. AI investments accelerated through acquisitions, including Scale AI, and Q3 guidance raised to $47.5-50.5B amid $66-72B capex plans for FY2025.

Microsoft's top engineers earn over $400k annually with new hire stock up to $2M, as tech giants compete fiercely for elite AI talent through high salaries, equity, and strategic acquisitions to drive AI dominance.

Poe launches API for seamless model integration, offering tiered credit-based plans starting at $4.99/month, supporting 100+ models including GPT-4o and Imagen 4, with free access via existing subscriptions and additional credit purchases.

Google DeepMind unveils AlphaEarth, an AI platform analyzing satellite data to track planetary changes, enabling efficient ecosystem monitoring and resource management via compact datasets and advanced geospatial models.