🚫 OpenAI blocked by Claude, 🧠 100 AI agents by Manus, 📚 ChatGPT Study Mode 🌍 AlphaEarth by Google, 🔐 GPT confidentiality warning

The most important news pieces from the AI world are here for you, whether you’re an AI enthusiast or just wish to keep an eye on what’s going on.

Focus Topic: Why enterprises are betting big on closed-source AI models

Based on insights from Menlo Ventures’ mid-2025 update to their report “The State of Generative AI in the Enterprise“

Foundation models are no longer just powering generative tools – they’re quietly reshaping the very infrastructure of enterprise computing. A new report from Menlo Ventures reveals a market shifting fast: in just six months, enterprise spending on large language model (LLM) APIs more than doubled, rising from $3.5 billion to $8.4 billion. This signals that AI in production is no longer an experiment – it’s the new baseline.

The most surprising result? Anthropic has overtaken OpenAI as the enterprise market leader, with 32% of workloads compared to OpenAI’s 25%. Their rise is no accident: the release of Claude Sonnet 3.5 and 3.7 introduced “agent-first” LLMs with real-world problem-solving power. Especially in code generation, Anthropic now holds a 42% market share – double that of OpenAI – and is driving the growth of tools like Claude Code, Cursor, and Windsurf.

Enterprise LLM market share by Menlo Ventures

But it’s not just model performance that’s changing the landscape. The way we scale intelligence is evolving, too. The future is no longer about bigger pretraining datasets – it’s about reinforcement learning with verifiers (RLVR) and agent-based tool usage. This shift is why 2025 is being called the year of agents – LLMs that think step-by-step, use calculators and search engines, and adapt in real time.

Perhaps most notably, open-source models are stalling. Despite advantages like customization and cost savings, enterprise adoption has dropped from 19% to 13%. The gap in performance, deployment complexity, and hesitation to rely on Chinese-origin models are all contributing factors. As a result, closed-source models aren’t just winning on capability – they’re becoming the default for serious AI builders.

LLMs & AI Models

    • Anthropic blocked OpenAI from using Claude, accusing it of violating terms during GPT-5 development with API-based comparison testing.
    • Manus AI launched a new Wide Research feature that runs 100 AI agents in parallel, simulating a personal research supercluster.
    • OpenAI introduced Study Mode in ChatGPT to guide students via Socratic questioning and step-by-step feedback, avoiding direct answers.
    • Z.ai released GLM-4.5, a 355B parameter open model excelling at logic, code, and agent tasks, positioned against o3 and Grok 4.
    • Google launched Deep Think for AI Ultra users – a faster, slightly less accurate version of its Olympiad-winning math model.
    • Microsoft added Copilot Mode to Edge browser, enabling voice commands, tab insights, and personal task support directly in-browser.

New Tools

    • FLUX.1 Krea is a new open-source image model by BFL and Krea, designed to eliminate the typical “AI look” and improve photorealism.
    • Amazon backed Showrunner, an “AI Netflix” platform letting users create and star in interactive animation series using text prompts.
    • NotebookLM added Video Overviews, allowing users to turn documents into narrated, visual presentations for learning or sharing.
    • Ideogram’s Character model maintains visual consistency of a person across scenes using just one reference photo for custom image generation.
    • Alibaba launched Wan2.2, a video model that surpasses Sora and Seedance in style, camera work, and text accuracy.

Wan2.2 demo video

    • Runway’s Aleph model enables text-driven video editing, allowing realistic changes to lighting, angles, scenes, and character details.
    • Figma Make is a new tool that turns static designs into interactive prototypes using simple text instructions.

👉 Explore these tools: FLUX.1 Krea | Showrunner | NotebookLM | Wan2.2 | Figma Make

Other Quick Picks

    • OpenAI will build Stargate Norway, its first European AI data center, powered by renewables and hosting 100,000 Nvidia GPUs.
    • Mark Zuckerberg shared Meta’s vision of personal AI assistants built into wearables, hinting future models may no longer be open.
    • Google DeepMind launched AlphaEarth, a virtual satellite model updating real-time maps and detecting environmental change.
    • Stanford unveiled a virtual AI science lab, where agents discovered COVID-19 nanobodies with 1% human input.
    • Meta tried to poach staff from Mira Murati’s startup with billion-dollar offers, but none accepted due to loyalty and belief in the vision.
    • Ethan Mollick questions whether AI needs to understand process or if outcomes alone are enough for business success.
    • China introduced an AI action plan calling for global collaboration, open source, and shared research infrastructure.
    • Sam Altman confirmed ChatGPT conversations lack legal confidentiality and could be used in court cases.

🇪🇪 AI News from EstAIonia

    • Thea Sogenbits outlined 5 questions to assess a company’s AI readiness, covering tools, data sensitivity, and human oversight.
    • Klemens Arro urges firms to treat AI as strategic infrastructure, not just chat tools, and to focus on real business value.
    • Estonian companies can now apply to join the €3B AI gigafactory project with Poland, Latvia, and Lithuania, contributing through investment or advisory roles.
    • Priit Kongo emphasized that today’s graduates enter a world where AI is essential for daily work, rewarding those who can guide it with creativity and critical thinking.

🎙️ AIPowerment Podcast was on summer holiday in July and will continue in August with new episodes.

🎧 Feel free to listen to existing AIPowerment Podcast episodes on Spotify, Apple Podcasts, and YouTube.

🙌🏼 If you found this article valuable, subscribe to weekly news updates.

💼 Exploring how to make AI work in finance? Let’s talk practical use cases – connect with Gerlyn Tiigemäe for expert guidance.

📚 If you would like to participate in one of my trainings or listen to speaking engagements, here are the upcoming ones:

Training “Kuidas endale luua personaalne AI assistant?”, 14.08.2025 – information

Related Posts

Scroll to Top