⏰ Limited-Time Free Models

Several free-tier models are nearing expiration on OpenRouter. NVIDIA Nemotron Nano 9B V2 and DeepSeek R1 Distill Llama 70B both expire tomorrow (June 11), priced at $0.04/$0.16 per M tokens and $0.70/$0.80 per M tokens respectively. Also ending soon: Z.ai GLM 4.5 (June 19, $0.60/$2.20 per M tokens) and Meta Llama 3 70B Instruct (June 19, $0.51/$0.74 per M tokens). Anthropic Claude Opus 4.6 Fast (1M context) remains free until June 29 at $30/$150 per M tokens.

🍎 Apple WWDC 2026: Siri Gets AI Supercharged by Gemini

Apple unveiled a dramatically overhauled Siri at WWDC 2026, powered by a $1 billion Google Gemini deal, alongside iOS 27 and a suite of new Apple Intelligence features. The new Siri AI runs on Google’s infrastructure, with Apple clarifying the assistant is “not Google Gemini with Apple branding” — its AI models contain none of Google’s Gemini Assistant code, though they leverage Gemini’s cloud compute. The announcement comes as Apple faces a $250 million false advertising lawsuit settlement over its earlier WWDC AI demonstrations that critics said overstated capabilities. Meanwhile, Apple’s strategy of betting on cheaper, on-device AI to woo small developers is drawing industry attention.

🎮 Anthropic Releases Claude Fable 5 — A Public Game-Making Version of Mythos

Anthropic has released Claude Fable 5, a public-facing version of the Claude Mythos architecture that can generate playable video games with the click of a button. TechCrunch reports Fable 5 makes “weirdly fun” games, from platformers to puzzle adventures, all generated through natural language prompts. The model is available today on OpenRouter with 1M context length at $10/M tokens prompt pricing. This marks Anthropic’s first consumer-friendly creative coding tool, positioning it against similar offerings from competitors.

🔬 Nvidia Partners with SK hynix, Clinches South Korea AI Deals

Nvidia announced a multiyear technology partnership with SK hynix to advance memory solutions for AI factories, and separately clinched deals with multiple South Korean giants to accelerate the AI boom. The partnerships cover HBM4 memory development and South Korea’s broader AI infrastructure push. In parallel, Nvidia CEO Jensen Huang declined an invitation to testify before the US Senate on AI, China, and export controls, adding to ongoing tensions between the chip giant and Washington policymakers.

💰 SoftBank Commits €75 Billion to Nuclear-Powered French Data Centers

SoftBank announced plans to invest up to €75 billion ($82 billion) to build nuclear-powered AI data centers in France — part of a wave of $108 billion in total foreign investment attracted by France. The data centers will be powered by nuclear energy, leveraging France’s abundant nuclear grid. SoftBank CEO Masayoshi Son also stated that “AI is designing OpenAI’s next model in a sign of superintelligence,” while reclaiming the top spot on Japan’s richest list amid the AI boom.

🏛️ Taiwan Weighs AI Chip Sales Curbs to China; California, Connecticut Enact AI Laws

Taiwan is exploring new restrictions on AI chip sales to China to align with US export controls, signaling a tightening of the global semiconductor supply chain. Domestically, California Governor Gavin Newsom signed a first-of-its-kind executive order to prepare workers and businesses for potential AI disruption, while Connecticut enacted comprehensive AI regulation that will impose new requirements on high-risk AI systems. The moves signal accelerating state-level AI governance in the absence of comprehensive federal legislation.

🚀 DeepSeek V4 Preview Sees Demand; Targets $59 Billion Valuation

DeepSeek’s V4 model preview, which TechCrunch reports “closes the gap with frontier models,” has driven surging adoption as US companies chase cheaper AI alternatives — the startup topped Ramp’s trending software vendors list in June. The company is now targeting a $59 billion valuation in its first-ever funding round, significantly higher than earlier $7 billion reports. The development intensifies the US-China AI rivalry, with the Council on Foreign Relations calling DeepSeek V4 “a new phase in the US-China AI rivalry.”

🔄 Google Cuts AI Plus Price; Transitions Gemini CLI to Antigravity

Google slashed the price of its AI Plus subscription plan and doubled included storage, firing a warning shot in the AI subscription price wars. Separately, the company announced it is transitioning the Gemini CLI to a new “Antigravity CLI,” signaling a platform-level shift in its developer tooling. Marvell also unveiled a 102.4Tbps AI-optimized switch targeting next-generation data centers.

In Brief

  • OpenAI confidentially filed for IPO, with the FT reporting plans for a ChatGPT “superapp” overhaul ahead of listing. Sam Altman’s eye-scanning company World (formerly Worldcoin) is reportedly doing layoffs.
  • Alibaba plans to integrate Qwen AI with Taobao for agentic shopping, while teasing new Qwen previews that rank as the highest-performing Chinese AI models on the Arena leaderboard.
  • Moonshot AI is moving to unwind its offshore structure in pursuit of an IPO, while Cursor admitted its new coding model was built on top of Moonshot’s Kimi architecture.
  • Super Micro announced a $7 billion capital raise to fund its AI backlog, causing its stock to drop 10%.
  • MCP ecosystem continues maturing: Pinterest deployed a production-scale MCP ecosystem for AI agent workflows, and Autodesk helped make MCP enterprise-ready.
  • Global AI spending is projected to grow 47% in 2026, per EE Times Asia.
  • Anthropic doubled Claude Cowork limits at no charge and continued its global expansion with India previews of Claude Mythos.

🤗 Trending on Hugging Face

Several notable models surfaced on Hugging Face today. unsloth/North-Mini-Code-1.0-GGUF brings efficient code generation to the GGUF format via Unsloth’s quantization. weathon/kto_review_qwen35_9b applies KTO alignment training to Qwen 3.5. Other interesting uploads include johannesmichalke/place-pencil-v1.1-rlt-online-v2 (RL-trained generation model) and Shamima/babylm-2026-multilingual-v3-quality-filter (data filtering for BabyLM). While HF activity is moderate today, the GGUF and alignment-focused models reflect ongoing trends in model efficiency.

🚀 New Models on OpenRouter

In addition to Claude Fable 5 (covered above), Nex AGI: Nex-N2-Pro is available on OpenRouter’s free tier with 262K context — a no-cost entry point for developers exploring long-context applications.

⭐ GitHub Trending: AI Edition

  • JimLiu/baoyu-design ★648 — Run Claude Design locally as an Agent Skill for Cursor, Claude Code, and more. Produce polished UI mockups from prompts.
  • GordenSun/GordenSuperPPTSkills ★640 — AI-powered PPT generation: generate luxury image-format PPTs and convert to fully editable PPTX files.
  • amElnagdy/guard-skills ★519 — Quality gates for AI-generated code: catch failure modes in tests, security, and logic during agent-driven development.
  • apple/coreai-models ★398 — Model export recipes, Python primitives, and Swift runtime utilities for on-device AI — coinciding with Apple’s WWDC AI push.
  • xiaohuailabs/xiaohu-video-translate ★361 — AI video translation pipeline: download, transcribe, translate, subtitle, and burn — all local, zero API fees.

💡 Key Trends

The platform wars intensify — Apple’s Gemini deal, Google’s subscription price cuts, and Anthropic’s Fable 5 release show AI companies competing fiercely for developers and consumers. Regulation accelerates at the state level — California, Connecticut, and Taiwan all advancing AI governance measures in a single day, filling the federal policy vacuum. Infrastructure spending reaches new heights — SoftBank’s €75B France bet, Super Micro’s $7B raise, and Marvell’s 102.4Tbps switch all point to a massive capex cycle that shows no signs of slowing. Apple finally enters the AI conversation — WWDC 2026 marks Apple’s most aggressive AI pivot yet, leveraging Google’s infrastructure while maintaining brand independence.