Daily AI News - 2026-05-30

⏰ Limited-Time Free Models

Several models on OpenRouter’s free tier are approaching expiration. The Gemini 2.0 Flash line ends its free run on June 1 (tomorrow!), while Qwen3 30B A3B, Llama 3 Euryale 70B v2.1, and Hermes 2 Pro Llama-3 8B all expire on June 5. The Claude Opus 4.6 Fast variant (1M context) is available until June 29.

🦁 Meta Launches Muse Spark, Abandons Open-Source Llama

Meta has unveiled Muse Spark, its first proprietary AI model since forming the Superintelligence Labs unit — marking a definitive break from the open-source Llama strategy that defined its AI identity for years (VentureBeat, Fortune, CNBC, Barron’s). The model arrives after Meta reportedly delayed its Llama successor, choosing instead to go closed-source amid an internal reorganization.

The shift is dramatic: Meta spent an estimated $14 billion on AI capex in Q1 2026 alone, and its stock surged 9% on the Muse Spark announcement. Critics including The Register and The New Stack have lambasted the move as Zuckerberg “abandoning open-source identity,” but investors rewarded the strategic pivot. Muse Spark represents Meta’s attempt to catch Google and OpenAI after spending billions — and signals that the era of free, open-source frontier models from Meta is over (CNBC, Barron’s).

🔍 Google Changes Its Search Box for the First Time in 25 Years

Google has fundamentally redesigned its search box for the first time in a quarter-century, shifting to an AI-first experience that blurs the line between search and conversational AI (NYT, TechCrunch, NPR, Axios, Google Blog). The new “Ask AI” mode sits alongside the classic search bar, offering AI-generated answers, proactive follow-ups, and contextual suggestions before users finish typing.

The change — announced at Google I/O 2026 and rolling out now — is described by TechCrunch as “Google Search as you know it is over.” The redesign has already sparked backlash: DuckDuckGo installs surged 30% as users object to being “force-fed Google’s AI Search” (TechCrunch). Google’s gamble is that AI integration will keep users within its ecosystem rather than turning to ChatGPT, Perplexity, or other AI-native search alternatives.

🏆 Anthropic’s $965B Valuation and the Mythos Fallout

Anthropic closed a $65 billion Series H at a $965 billion post-money valuation, leapfrogging OpenAI to become the world’s most valuable AI startup (NYT, Bloomberg, Reuters). The round — the largest single fundraising in tech history — drew Samsung and SK Hynix as first-time chipmaker investors (KED Global).

Meanwhile, the security implications of Claude Mythos are dominating headlines. The World Economic Forum called it a “moment for cybersecurity,” as researchers project AI-boosted cyberattacks could cost banks billions (Reuters). Anthropic released Project Glasswing, an initiative to secure critical software for the AI era, while CEO Dario Amodei warned of a cyber “moment of danger” with thousands of newly exposed vulnerabilities (CNBC, Forrester).

🚀 OpenAI: GPT-5.5 Instant, Erdős Breakthrough, and Rosalind Biodefense

OpenAI has rolled out GPT-5.5 Instant as the new default model for ChatGPT, promising more accurate, concise, and personalized responses with reduced hallucinations and deeper memory (TechCrunch, Axios, Mashable, SiliconANGLE). The upgrade is available to all users now.

In a stunning scientific milestone, an OpenAI model autonomously disproved an 80-year-old Erdős conjecture in discrete geometry — a problem that had resisted human mathematicians since the 1940s (OpenAI, TechCrunch, Scientific American). In a dramatic twist, Anthropic’s Claude Mythos reportedly produced a simpler, more elegant proof for the same result (the-decoder.com).

OpenAI also launched Rosalind Biodefense, offering its life sciences AI model free to governments for pandemic preparedness and biodefense (OpenAI, Axios). The company briefed the White House on the model, which is designed to accelerate vaccine research and detect emerging biological threats.

🔌 OpenRouter Hits $1.3B Valuation with $113M Series B

OpenRouter, the AI model routing platform, has raised $113 million in Series B funding led by Google’s CapitalG and joined by NVIDIA — more than doubling its valuation to $1.3 billion in a year (TechCrunch, SiliconANGLE, PYMNTS.com). The platform now processes 25 trillion tokens weekly across 400+ models.

The funding round underscores the exploding demand for multi-model inference infrastructure. OpenRouter acts as a unified marketplace and routing layer, letting developers access models from OpenAI, Anthropic, Google, Meta, and dozens of smaller labs through a single API — a model that has resonated as the AI ecosystem fragments across competing providers and pricing tiers.

🏭 Mistral Expands Industrial AI with Airbus, BMW, and Emmi AI Acquisition

Mistral AI has signed major contracts with Airbus and BMW for on-premise AI deployments in aerospace and manufacturing (Bloomberg, Airbus official). The French AI company is positioning itself as Europe’s sovereign AI champion, offering fully on-premise deployments that comply with EU data regulations.

Mistral also acquired Emmi AI, a physics AI startup, to expand into engineering simulation and industrial modeling (HPCwire). The company launched a Search Toolkit for enterprise knowledge retrieval and announced its Copilot Studio integration with Microsoft. The flurry of activity signals Mistral’s ambition to become the full-stack AI provider for European industry — from foundational models to domain-specific applications.

🤖 xAI Launches Grok Build 0.1 Coding Agent

xAI has launched Grok Build 0.1, its first AI coding agent, entering the increasingly crowded AI coding assistant race alongside Claude Code, GitHub Copilot, and Codex (PCMag, xAI official). The agent is available in early beta via API and CLI, with Windows PowerShell support.

However, adoption faces headwinds: Reuters reports that “Grok falls flat in Washington,” undermining SpaceX’s AI growth story with government clients. A class-action suit has also been filed over Grok’s alleged “undressing” controversy (Tech Policy Press). Despite these challenges, xAI is pressing ahead, with a larger coding model reportedly targeting a mid-June release.

🏗️ SoftBank Plans €75B French AI Data Center Investment

SoftBank has announced plans to invest up to €75 billion ($87.5B) to build AI data centers in France — Europe’s largest AI infrastructure project (Fortune, Reuters, TechCrunch, FT, WSJ). The investment, unveiled during a meeting between Masayoshi Son and President Macron, would create thousands of jobs and position France as a European AI compute hub.

The scale is staggering: SoftBank’s commitment rivals the GDP of mid-sized economies. The move comes amid a broader AI infrastructure boom — Goldman Sachs estimates $800 billion in AI capex will flow in 2026, with Amazon alone planning $200 billion (Bloomberg, Goldman Sachs, The Motley Fool). ByteDance is reportedly weighing up to $70 billion in AI capex (Bloomberg).

💸 DeepSeek Makes 75% V4-Pro Price Cut Permanent

DeepSeek has made its 75% price reduction on V4-Pro permanent, cementing its position as the pricing disrupter in the LLM market (VentureBeat, Engadget, InfoWorld). The Chinese AI lab’s flagship model now costs $0.10/M tokens for input and $0.30/M for output — roughly 34x cheaper than GPT-5.5 on output tokens.

The permanent cut signals DeepSeek’s strategy of buying market share through aggressive pricing. Enterprise adoption of V4-Pro surged 3x since the initial rollout (Semafor). Meanwhile, China has reportedly expanded travel curbs on top AI talent at private firms (Bloomberg), signaling growing state control over the flow of AI expertise.

🦾 Alibaba’s Qwen 3.7 Max Runs 35 Hours Autonomously

Alibaba unveiled Qwen3.7-Max, a new agentic AI model that can run autonomously for 35 hours, executing over 1,000 tool calls before requiring human intervention (VentureBeat, SCMP, the-decoder). The model supports external harnesses including Anthropic’s Claude Code, positioning it as a flexible cross-platform automation agent.

Alibaba also debuted a more powerful Zhenwu AI chip at its Singapore Cloud Summit (CNBC), and announced integration of Qwen AI with Taobao for agentic shopping experiences (Reuters). The moves cement Alibaba’s ambition to become China’s integrated AI factory — providing compute, models, and applications as a unified stack.

🪟 Microsoft Builds ‘One Copilot’ Super App

Microsoft is building a unified “One Copilot” super app that combines GitHub Copilot coding, AI chat, and agentic tools into a single platform (Fortune, Benzinga). The super app consolidates Microsoft’s sprawling AI product line — currently spread across Copilot, GitHub Copilot, Microsoft 365 Copilot, and Azure AI Studio — into one cohesive experience.

CEO Satya Nadella acknowledged the “Copilot confusion” problem in an internal town hall, where employees laughed at the proliferation of branded AI tools. Separately, Mistral’s models are joining Copilot Studio’s lineup, and Microsoft is developing a proprietary AI coding model to reduce reliance on OpenAI (Cybernews), signaling a strategic pivot toward greater independence in the AI stack.

🚀 New Models on OpenRouter

One new model appeared on OpenRouter:

• StepFun: Step 3.7 Flash — $0.20/M tokens prompt, 256K context. StepFun’s latest efficient flash model, optimized for speed and cost at a fraction of flagship pricing.

⭐ GitHub Trending: AI Edition

🤗 Trending on Hugging Face

Hugging Face activity remained quiet today. The latest uploads consist mostly of zero-download experimental checkpoints and region-tagged placeholder models. No major model releases from established labs appeared on the platform.

💡 Key Trends

1. The open-source retreat: Meta’s abandonment of open-source Llama for proprietary Muse Spark signals that even the most vocal open-source advocates are retreating behind closed doors as the AI arms race intensifies.
2. Infrastructure escalation: SoftBank’s €75B France commitment, Amazon’s $200B capex, and Goldman’s $800B total estimate — AI infrastructure spending is entering a phase that rivals national budgets.
3. The routing layer emerges: OpenRouter’s $1.3B valuation reflects a growing realization that the AI ecosystem needs middleware — a model-agnostic routing layer that handles the complexity of 400+ competing providers.
4. AI-for-science accelerates: OpenAI’s Erdős problem solution, the Rosalind Biodefense program, and the independent Claude Mythos proof all point to AI moving from text generation to genuine scientific contribution.