August 2025 AI Roundup

OpenAI launches GPT-5, a unified system that combines multiple models - a fast model for quick responses and a deeper reasoning model for complex problems. A smart router automatically decides which model to use based on your question's complexity. Available to all users, with Plus and Pro subscribers getting higher usage limits and access to extended reasoning.

Anthropic releases Claude Opus 4.1, an upgrade to Claude Opus 4 that improves coding, reasoning, and research tasks. The model scores 74.5% on the SWE-bench Verified coding benchmark and shows significant improvements in multi-file code refactoring and data analysis.

DeepMind announces Genie 3, an advanced AI that generates interactive 3D worlds from text descriptions, maintaining visual consistency for several minutes. Users can explore realistic environments like volcanic landscapes, underwater scenes, and fantasy realms, plus modify them in real-time through text commands like changing weather or adding objects.

ChatGPT Agent combines OpenAI's previous tools (Operator and Deep Research) into one unified system that can autonomously handle complex workflows using its own virtual computer. It navigates websites, conducts research, creates presentations, and manages tasks like calendar scheduling and competitor analysis, all while keeping users in control through natural language prompts.

Midjourney launches HD Mode for AI video, delivering 720p quality with improved motion and fewer visual artifacts. The feature costs 3.2x more than standard video but is exclusive to Pro and Mega subscribers, targeting professionals who need higher-quality output without external upscaling tools.

Elon Musk's xAI has launched Grok Imagine, an AI image and video generator for SuperGrok & X Premium+ subscribers. The tool creates 15-second videos from text prompts and can animate static images, positioning itself as a less restricted alternative to competitors like Google and OpenAI, though current video quality remains somewhat cartoonish.

Google DeepMind and OpenAI's AI models each solved 5 out of 6 problems in the International Mathematical Olympiad, achieving gold medal-level performance for the first time. Unlike previous AI attempts, these systems worked entirely in natural language, bringing AI closer to solving complex real-world scientific problems.

Google launches Deep Think in Gemini app for AI Ultra subscribers, using parallel thinking techniques to tackle complex problems in coding, math, and research. The new feature delivers more detailed responses while being faster and more practical than Google's gold-medal math competition model.

Both OpenAI and Google released new AI learning features - ChatGPT Study Mode and Guided Learning in Gemini - that guide students through problems step-by-step instead of giving direct answers. The tools use interactive prompts, personalized support, and research-backed teaching methods to help students build deeper understanding rather than just complete assignments.

OpenAI releases gpt-oss-120b and gpt-oss-20b, its first open-weight language models since GPT-2, under Apache 2.0 license. These cost-effective models excel at reasoning, tool use, and run efficiently on consumer hardware, making advanced AI accessible to all developers.

Our comprehensive Midjourney analysis reveals how the platform became the leading AI image generator, with explosive revenue growth and a massive global community. The report covers market dominance, user demographics, traffic patterns, and what's driving its success against competitors.

AIPRM now works with Claude Opus 4.1 and GPT-5 models. We're always updating both AIPRM for ChatGPT and AIPRM for Claude to stay current with the newest changes, and we recommend retesting your own custom prompts to get the best results from these improved models.