⏰ Limited-Time Free Models
Several free LLM tiers are expiring soon on OpenRouter:
| Model | Expires | Pricing |
|---|---|---|
| Mistral 7B Instruct v0.1 | May 30 (1 day) | Free |
| Xiaomi MiMo-V2-Omni | May 31 (2 days) | Free |
| Xiaomi MiMo-V2-Pro | May 31 (2 days) | Free |
| Gemini 2.0 Flash Lite | Jun 1 (3 days) | Free |
| Gemini 2.0 Flash | Jun 1 (3 days) | Free |
| Qwen3 30B A3B | Jun 5 (6 days) | Free |
| Llama 3 Euryale 70B v2.1 | Jun 5 (6 days) | Free |
| Hermes 2 Pro Llama-3 8B | Jun 5 (6 days) | Free |
🚀 Anthropic Hits $965B Valuation, Launches Claude Opus 4.8
Anthropic has closed a $65 billion Series H funding round at a post-money valuation of $965 billion, surpassing OpenAI to become the world’s most valuable AI startup (NYT, Bloomberg, Reuters). The company also announced Claude Opus 4.8, its latest flagship model now available on OpenRouter at $10/M tokens input (fast variant) and $5/M tokens input (standard), with 1M token context windows. Opus 4.8 introduces a new “dynamic workflow” tool that enables multi-step autonomous task execution, and early benchmarks show it leaves four times fewer code flaws unflagged than GPT-5.5 (TechCrunch, ZDNET). Meanwhile, Anthropic confirmed its upcoming Claude Mythos model family is “weeks away” from release — a model so powerful it has already raised cybersecurity concerns and prompted EU regulators to seek access (Reuters, Decrypt, Bloomberg). The bumper funding and product acceleration position Anthropic for a potential IPO later this year (TechCrunch).
💰 DeepSeek Makes 75% V4-Pro Price Cut Permanent
China’s DeepSeek has permanently slashed pricing on its flagship V4-Pro model by 75%, igniting a fresh round of the global AI pricing war (Reuters, Bloomberg, Engadget). The move comes as DeepSeek V4 Pro now tops global “bang-for-buck” rankings according to independent benchmarks from the South China Morning Post. In an even more aggressive move, Xiaomi has joined the price war, slashing its own API prices by up to 99% for direct competition with DeepSeek (36Kr). The cuts reflect a broader market trend: as frontier model capabilities converge, pricing has become the primary differentiator — and Chinese AI labs are winning that race decisively.
🏛️ White House Considers Vetting AI Models Before Release
The Trump administration is moving toward mandatory pre-release government review of major AI models, according to the New York Times, Reuters, and CNBC. The proposal would require companies like Google, Microsoft, xAI, and OpenAI to submit new frontier models for federal testing before public deployment. This represents a significant escalation of federal AI oversight — a sharp departure from the administration’s earlier hands-off approach. However, internal White House divisions remain: Politico reports that the security hawks pushing for review are clashing with free-market advisors who warn the policy could hand China a competitive advantage. The NYT reports that the first models to be tested would include those from Google, Microsoft, and xAI.
🛠️ xAI Enters Coding Agent Race with Grok Build
Elon Musk’s xAI has launched Grok Build, its first AI coding agent, now in early beta for X Premium+ subscribers (Engadget, PCMag, DevOps.com). Grok Build competes directly with Claude Code, OpenAI Codex, and Cursor — and is available via terminal on macOS. The launch positions xAI as a serious contender in the developer tools space, leveraging the Grok ecosystem and xAI’s compute partnership with Anthropic (reportedly $1.25B/month in compute trades, per TechCrunch). Grok Build marks xAI’s pivot from consumer chatbot to developer infrastructure.
💸 AI Costs Begin to Bite: Goldman Sachs Reports 24x Token Demand Spike
A major Goldman Sachs report warns that AI agents could increase token demand by 24x, with real-world costs already shocking enterprise adopters (Tom’s Hardware, Goldman Sachs). Uber burned through its entire 2026 AI budget in just four months, with its COO publicly questioning whether the spend is worth it (Fortune). Microsoft internal reports show that in some deployments, AI compute costs exceed the salaries of the human employees it was meant to replace (Fortune). An NVIDIA executive acknowledged the paradox, stating “the cost of compute is far beyond the costs of the employees.” The reports come as Goldman Sachs lifts its 2026 AI capex estimate to $800 billion, with expectations of $1.1T by 2027 (Benzinga, Yahoo Finance).
🤖 Microsoft Build 2026 Previews AI-First Windows
Microsoft Build 2026 is previewing the most aggressive AI integration yet — computer-using agents, real-time voice experiences, and a new workflows platform that effectively turns Windows into an AI operating system (PCMag, Microsoft). The company demonstrated agents capable of controlling desktop applications autonomously, building on its “Windows Copilot” vision. Meanwhile, Microsoft’s stock rose 3.57% on May 28, with analysts suggesting AI revenue may be significantly underestimated (Investing.com, TradingView). The Build conference will also feature Windows 12 previews, with AI deeply embedded at the kernel level (TechRadar).
🇨🇳 Alibaba Unveils Qwen 3.7 Max: 35-Hour Autonomous Runtime
Alibaba has released Qwen 3.7 Max, a proprietary AI model capable of running autonomously for 35 hours and making over 1,000 tool calls without human intervention (VentureBeat, Pandaily, Gigazine). The model supports external harnesses including Anthropic’s Claude Code and has quietly outperformed Opus 4.7 in coding benchmarks (Geeky Gadgets). Separately, Alibaba unveiled its Zhenwu M890 AI chip, advancing its push for domestic semiconductor alternatives amid ongoing US export restrictions (Reuters, Quartz). The company also announced plans to integrate Qwen AI with Taobao for agentic shopping experiences (Reuters, KrASIA).
🔒 OpenAI Releases Frontier Governance Framework
OpenAI has published its Frontier Governance Framework, a new policy blueprint outlining how the company plans to manage risks associated with increasingly powerful AI models (OpenAI blog). The framework addresses pre-deployment testing, transparency requirements, and emergency shutdown protocols. Separately, the OpenAI Foundation committed $250 million to studying AI’s impact on jobs and communities (Staffing Industry Analysts), and Accenture Federal Services partnered with OpenAI to accelerate secure AI adoption across the US federal government (Accenture).
🤗 Trending on Hugging Face
| Model | Downloads | Notes |
|---|---|---|
| NDIF/hackathon-imposter-syndrome-eve-llama8B-v2 | 40 | LoRA adapter on Llama 3.1 8B |
| E2M/yasa | New (1 like) | GGUF quantized text-gen model |
| MediMind-411M | New | Medical LLM with PyTorch |
The HuggingFace ecosystem sees quiet activity today, with most uploads being experimental or zero-download submissions. The NDIF hackathon model garnered the most traction among recent uploads.
⭐ GitHub Trending: AI Edition
| Repo | Stars | Description |
|---|---|---|
| OpenBMB/PilotDeck | 1,451 | Task-oriented AI Agent productivity platform |
| study8677/awesome-architecture | 722 | 21 architecture maps incl. AI gateway, RAG, agents |
| UditAkhourii/adhd | 460 | Tree-of-thought coding agent skill for Claude SDK |
| FlashML-org/flashlib | 356 | Fast and memory-efficient classical ML operators |
| VILA-Lab/FigMirror | 354 | AI tool to plot data in any paper’s figure style |
💡 Key Trends
- Valuation arms race peaks: Anthropic’s $965B valuation — surpassing OpenAI — signals that AI infrastructure is becoming the most capital-intensive sector in tech history, with an IPO wave approaching.
- AI’s cost paradox deepens: Despite Goldman’s $800B capex estimate, enterprise deployment costs (Uber’s 4-month budget burn, Microsoft’s compute-vs-salary math) reveal that AI’s unit economics remain unresolved at scale.
- Regulation accelerates on two fronts: The White House’s pre-release vetting proposal and state-level accountability bills (Illinois, Colorado) show AI governance shifting from voluntary to mandatory — on both sides of the Atlantic.