OpenAI's $110B War Chest, Anthropic's Political Stand, and Google Ships Three
March 2, 2026
It was a week of contrasts. OpenAI closed the largest startup fundraise in history — $110B at an $840B valuation — while Anthropic made headlines for pushing back on the Secretary of War. Meanwhile, Google quietly shipped Gemini 3.1 Pro, a new image model, and music generation all at once.
Today: The mega-round, Claude's dominance streak, Google's triple drop, and why your agentic benchmarks might be lying to you.
🧠 MODEL RELEASES
Gemini 3.1 Pro lands. Google's latest is built for complex, multi-step reasoning — not quick answers. Designed for tasks where depth matters more than speed. Read more →
Nano Banana 2 ships at Flash speed. DeepMind's image generation model combines Pro-level capabilities with lightning-fast inference — advanced world knowledge, subject consistency, and production-ready specs. Read more →
Qwen3.5-35B-A3B drops. Alibaba's new mixture-of-experts model packs 35B parameters but only activates 3B at inference — multimodal image-text with 680K+ downloads already on HuggingFace. Read more →
🔧 DEVELOPER & PLATFORM
Claude Code gets /simplify and /batch. Two new skills that automate shepherding PRs to production and parallelizing code migrations. The Claude Code team says they've been using both daily. (12,011 likes) Read more →
Claude Code Remote rolls out for Pro users. Run Claude Code from anywhere — no local setup required. (3,471 likes) Read more →
Claude Code's AskUserQuestion now renders markdown. Diagrams, code snippets, and rich content directly in the approval flow. (4,524 likes) Read more →
Bun v1.3.10 fixes 155 issues. Native JS/TS REPL, Windows ARM64 support, ECMAScript decorators, and 642 upvoted issues addressed in one release. Read more →
📝 RESEARCH & INSIGHTS
A team of parallel Claudes built a C compiler. Anthropic tasked Opus 4.6 agent teams to build one autonomously — then mostly walked away. The results reveal a lot about where autonomous software development is heading. Read more →
Your agentic benchmarks are probably wrong. Infrastructure configuration can swing agentic coding eval scores by several percentage points — sometimes more than the gap between top models on the leaderboard. Read more →
Demystifying evals for AI agents. The capabilities that make agents useful also make them hard to evaluate. New strategies for matching eval complexity to system complexity. Read more →
Inside the M4 Apple Neural Engine. A deep reverse-engineering teardown of Apple's latest neural hardware — Part 1 of what promises to be a fascinating series. (330 likes) Read more →
📱 PRODUCTS & ECOSYSTEM
OpenAI closes $110B from Amazon, NVIDIA, and SoftBank. The largest startup fundraise in history, at an $840B post-money valuation. The AI arms race now has a price tag. Read more →
Claude hits #1 in the App Store. Anthropic's consumer moment is real. (6,245 likes) Read more →
Meta's AI smart glasses raise privacy alarms. Workers say "we see everything" — and regulators are starting to listen. (1,034 likes) Read more →
🔥 TRENDING
🔥 Anthropic pushes back on the Secretary of War. The company issued a public statement responding to comments from Pete Hegseth — and the internet overwhelmingly approved. (42,396 likes | 6,713 RTs) Read more →
🎵 Gemini can now make music. Google's Lyria 3 model lets anyone create 30-second tracks from text or images, directly in the Gemini app. Read more →
💸 What ad-supported AI chat looks like. Someone built a working demo — and it's exactly as dystopian as you'd expect. (476 likes) Read more →
⚡ QUICK LINKS
- Why We Think: Lilian Weng's deep dive on test-time compute and chain-of-thought — why "thinking time" works. Link
- DeepSeek-R1: Still trending on HuggingFace with 928K downloads and 13K likes. Link
- prompts.chat: Community prompt library hits 149K stars — now self-hostable with full privacy. Link
Until next time ✌️