AI News This Week: AI model releases July 2026 | AI Weekly Pulse #4

Moonshot AI just released Kimi K3, a 2.8-trillion-parameter sparse Mixture-of-Experts model that, at the time of release, is the largest open AI model in the world. It handles text, images, and video natively, supports a 1-million-token context window, and uses MXFP4 weight quantization to bring storage down to roughly 1.4 TB - manageable enough for multi-node self-hosting. Two new architectural additions, Kimi Delta Attention (KDA) and Attention Residuals, handle the long-context efficiency side of things, while a LatentMoE framework coordinates 896 experts, with about 16 active at any given time. The model is live now through the Kimi API and apps, with full open weights coming July 27, 2026. Pricing sits at around $3 per million input tokens and $15 per million output tokens - noticeably cheaper than comparable U.S. frontier models. On benchmarks, it took first place across six of seven domains in the Frontend Code Arena and scored 88.3 on Terminal-Bench 2.1.

Thinking Machines Lab shipped Inkling this week - the first model the company has trained from scratch. It's a 975B-parameter Mixture-of-Experts model with 41B parameters active at inference time, pretrained on 45 trillion multimodal tokens. Text, image, and audio inputs are all supported natively, the context window stretches to 1 million tokens, and the whole thing is released under Apache 2.0, so the weights are yours to download from Hugging Face and do with as you please. A lighter Inkling-Small preview, running 12B active parameters, came out alongside it. Both are fine-tunable through Thinking Machines' own Tinker platform and work with multiple inference engines. The intent here isn't to build another chatbot - it's to give enterprises a serious foundation to customize on top of. That positioning also makes it one of the more substantial Western open-weights releases during a stretch when Chinese labs have been doing most of the heavy lifting.

France's competition authority just dropped a 3,700-page advisory opinion on the AI agent market, and the headline number is hard to ignore: OpenAI, Google, and Anthropic together control more than 84% of it. The inquiry started in January 2026 and took an unusually hands-on approach - regulators didn't just read reports, they built and deployed their own AI agents, running them through 550 shopping-related prompts to see how these systems actually behave in the wild. What they found was a market that has quietly shifted from simple chatbots to autonomous agents capable of reasoning through problems and coordinating real-world actions. The authority's concern is that this shift, left unchecked, could pull the broader digital economy into the orbit of a small number of vertically integrated tech giants. The report calls for regulatory intervention on three fronts: data access, interoperability between AI systems, and how default AI placements are handled.

Anthropic, Blackstone, Hellman & Friedman, and a group of supporting investors have launched Ode with Anthropic, a standalone enterprise AI services company built on the acquisition of Fractional AI. The idea is straightforward: embed Anthropic engineers and Claude models directly into mid-to-large enterprises to help them actually build, deploy, and maintain production AI systems - not just prototype them. The focus is on the unglamorous but necessary work: integrating AI into existing operations, navigating compliance requirements, and handling the complex workflows that most off-the-shelf solutions weren't designed for. The Fractional AI founders are leading it.

Meta pulled its "Muse" AI image generation feature from Instagram within days of launching it, after users pushed back hard on how the tool handled privacy and consent. The mechanics were simple and the problem was obvious in hindsight: anyone could generate AI images using photos from any public Instagram account just by tagging it. Every public account was opted in by default. No notifications went out when someone's photos - or likeness - were used. Meta's public response was that it had "missed the mark," which is one way to put it.

Databricks closed a strategic funding round led by Coatue, raising roughly $3 billion and pushing its valuation to $188 billion - up from $134 billion previously. The money is going toward accelerating three specific product areas: Unity AI Gateway, which handles governance across multiple AI systems; Genie, which connects data to automated actions; and Lakebase, built for AI agent workloads.