The AI landscape has been remarkably active over the past 72 hours, with major model launches from both established players and high-profile startups, a blockbuster compute deal, China asserting itself on the global governance stage, and a flurry of open-source milestones. Here is your daily briefing for July 18, 2026.
Model Releases
Thinking Machines Lab Debuts Inkling — A 975B Open-Weights Challenger. On July 15, Thinking Machines Lab — the AI startup founded by former OpenAI CTO Mira Murati and backed by Nvidia at a $12 billion valuation — released its first in-house model, Inkling. The 975-billion-parameter open-weights model was trained on multimodal data including video and audio, and is positioned as a customizable alternative to monolithic frontier systems. Unlike the “one-size-fits-all” approach of many labs, Thinking Machines is betting that developers want fine-grained control over model behavior. The release marks Murati’s first major product since leaving OpenAI and signals that the open-weights race now has a credible new entrant with serious backing (TechCrunch, WIRED).
Moonshot AI Drops Kimi K3 — The Largest Open-Source Model Ever. Just one day later, on July 16, Beijing-based Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter Mixture-of-Experts model that the company claims rivals — and on some benchmarks surpasses — OpenAI’s GPT-5.6 and Anthropic’s Claude Sonnet 5. Built with novel architectural innovations including Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), K3 offers a 1-million-token context window with native vision and always-on reasoning. Moonshot says full model weights will be released under an open-source license on July 27, making it by far the largest open-weight model ever shipped. Early benchmarks show K3 topping the Artificial Analysis Intelligence Index and the Chatbot Arena leaderboard (CNBC, VentureBeat, BBC).
Industry Moves
xAI Officially Becomes SpaceXAI. On July 6, Elon Musk’s AI company completed its rebrand to SpaceXAI, retiring the xAI name after the subsidiary was fully absorbed into SpaceX. The move follows SpaceX’s blockbuster IPO in June and Musk’s earlier announcement that Grok and X’s AI division would operate under the SpaceX umbrella. The rebrand positions SpaceXAI as a vertically integrated AI infrastructure play, combining rocket-based connectivity, the Colossus 2 data center in Memphis, and Nvidia’s latest hardware — a strategy Musk argues is necessary given that terrestrial electricity infrastructure alone cannot meet future AI demand (Business Insider, Wikipedia).
SpaceX Inks $6.3B Compute Deal with Reflection AI. In a massive bet on open-source AI infrastructure, SpaceX signed a $150-million-per-month compute deal with Reflection AI starting July 1, 2026, running through 2029. The agreement gives Reflection dedicated access to Nvidia GB300 chips inside SpaceX’s Colossus 2 data center near Memphis, Tennessee. Combined with a separate $1 billion deal Reflection signed with Nebius, the startup’s total compute commitments now exceed $7 billion — an extraordinary burn rate for a company whose open-source ethos has made it a darling of the developer community (TechCrunch, Benzinga).
ZTE Unveils the World’s First “Agentic AI” Smartphone. At the World AI Conference in Shanghai, ZTE debuted the NaviX Ultra, which it bills as the first smartphone with built-in “agent-like” AI capabilities. Rather than simply layering a chatbot on top of a mobile OS, the device embeds AI agents directly into the system layer, capable of executing multi-step tasks on the user’s behalf. The launch underscores how Chinese manufacturers are racing to redefine the smartphone form factor for the AI era, with IDC forecasting that AI-enabled phones will account for more than half of China’s market by the end of 2026 (Bloomberg, Yahoo Finance).
Funding Rounds
AI Agent Startups Raise $1.8B in July Alone. AI agent startups — companies building autonomous systems that execute real-world tasks — collectively raised $1.8 billion across more than a dozen deals so far in July 2026, according to sector tracker AIFunding.me. Sequoia Capital, Index Ventures, and Andreessen Horowitz led the deal flow, while average valuations climbed 40% quarter-over-quarter to $280 million. The surge reflects a market that increasingly sees AI agents as the next application layer beyond chatbots, with enterprise automation and developer tooling attracting the largest checks (AI Funding, Spearhead).
Reflection AI’s Compute Gambit Tops $7B. While not a traditional equity round, Reflection AI’s back-to-back compute deals — $6.3 billion with SpaceX and $1 billion with Nebius — represent the largest infrastructure commitment ever made by an open-source AI company. The deals effectively function as a massive capital raise, committing Reflection to $150 million per month in compute costs through 2029 in exchange for dedicated access to cutting-edge Nvidia hardware. The strategy reflects a growing trend: as frontier model training costs soar, compute-access agreements are becoming the new funding mechanism for AI startups (The Next Web, The Technology Express).
Open-Source News
Mozilla’s State of Open Source AI Report: 2.5 Million Models, 13 Million Builders. Mozilla published its inaugural State of Open Source AI report on July 14, painting a picture of a movement that has reached critical mass. The report, based on a global survey of 950+ developers and fresh ecosystem analysis, found 2.5 million open models now available, with 13 million builders actively contributing. Crucially, 79% of developers adding AI functionality to their products now use open models, compared to 71% for proprietary alternatives — signaling a pendulum swing toward openness. However, the report also identified a stark revenue gap: open-source AI captures only 4% of industry revenue, with the bottleneck being operational deployment costs rather than model quality (Mozilla Blog, TIME).
Kimi K3 Sets a New Open-Weight Record. Moonshot AI’s Kimi K3 isn’t just another model release — the decision to open-source 2.8 trillion parameters reshapes the competitive landscape for open-weight AI. By comparison, DeepSeek’s V4, previously the largest open model at 1.6 trillion parameters, is now dwarfed by nearly double. The full release on July 27 will give the global developer community access to a model that, on several benchmarks, currently sits at or near the top of frontier rankings. This is a watershed moment for the open-source AI movement, demonstrating that openness need not come at the cost of frontier capability (New York Times, Silicon Angle).
Policy & Regulation
Xi Jinping Launches China’s Rival AI Governance Bloc at WAIC 2026. On July 17, addressing the World AI Conference in Shanghai — his first-ever WAIC keynote — Chinese President Xi Jinping formally launched a new inter-governmental AI alliance, positioning China as the champion of open, cooperative AI governance in explicit contrast to what Beijing frames as U.S. hegemonic control. The alliance, formed on July 16, aims to develop shared AI regulations across participating countries and promote “human-centered” AI development. UN Secretary-General António Guterres appeared via video link, warning against AI monopoly concentration. The conference, running through July 20, features over 300 product premieres and major announcements from Huawei, Agibot, and Unitree (Al Jazeera, Tech Times, Reuters).
EU AI Act’s August 2 Deadline Looms — Despite Omnibus Delay. While the EU’s May 2026 Omnibus amendment pushed most high-risk AI system compliance obligations to December 2027, the August 2, 2026 deadline still activates crucial provisions: General-Purpose AI (GPAI) transparency and copyright rules, Article 50 disclosure obligations for AI-generated content, and the requirement for every member state to have at least one operational AI regulatory sandbox. The staggered enforcement creates a complex compliance landscape where some rules are already in effect, others kick in on August 2, and the most stringent high-risk obligations remain two years away. With fines reaching up to €35 million or 7% of global revenue, enterprises operating in the EU face a fragmented but urgent compliance calendar (DLA Piper, Informed Clearly, State of Surveillance).
Andy Burnham’s Plan to Scrap UK Tech Department Triggers Backlash. Incoming UK Prime Minister Andy Burnham — who became Labour leader this week after Keir Starmer’s resignation — faces mounting criticism over his plan to dismantle the Department for Science, Innovation and Technology (DSIT). MPs and industry leaders warn that reorganizing Whitehall at a critical moment for AI competitiveness would waste time and resources. Burnham’s team has signaled a shift toward a more interventionist, sovereignty-focused tech strategy that deprioritizes the US-centric model in favor of domestic British AI champions and stronger worker protections against AI-driven displacement (The Guardian, Chatham House).
See how it compares → /comparisons/
📖 Related Reads
- NiteAgent — AI agent development, frameworks, and production patterns
- CodeIntel Log — code quality, debugging, and software engineering benchmarks
- ToolBrain — tool reviews, LLM comparisons, and AI workflow guides
Cross-links automatically generated from None.
Back to all posts
