2026-03-22
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• GPT-5.4 mini goes free — paid model tiers start looking like a relic
• Gemini 3.1 Flash-Lite at $0.25/M changes what’s economically buildable
• Apple’s AI Siri lands on a billion devices — distribution is the new moat
• OpenAI at $25B ARR, eyes IPO — public markets will test every AI valuation
• GitHub Copilot Workspace closes the issue-to-PR loop automatically
Something shifted today. OpenAI rolled out GPT-5.4 mini to all ChatGPT Free and Go users via the Thinking feature — and retired GPT-5 Thinking mini with a 30-day countdown. That’s three years of AI pricing pressure reaching its logical conclusion: what cost $20/month is now free. The capability race is over. The distribution race has begun.
Why it matters: When frontier reasoning is free, the moat stops being the model and starts being everything around it — the interface, the context, the integration. OpenAI knows this. Every other lab knows it too. Watch what they build next, not what they release.
Google launched Gemini 3.1 Flash-Lite with 2.5× faster response times and 45% faster output, priced at $0.25 per million input tokens. On the Artificial Analysis Intelligence Index, Gemini 3.1 Pro now ties GPT-5.4 Pro at 57 points — the benchmark gap has closed to a rounding error. At $0.25/M, high-volume AI applications stop being an infrastructure question and become a product question. That threshold matters more than the ranking.
Apple shipped the AI-native Siri with iOS 26.4 today, completing a multi-year rebuild. The new assistant integrates on-device context with third-party apps natively. This is the largest single AI distribution event in history — not because of the technology, but because of the install base. A billion users’ first real AI experience is about to be shaped by Apple. That’s a gravitational shift no benchmark measures.
OpenAI has crossed $25 billion in annualized revenue and is taking early steps toward a public listing, potentially late 2026. Anthropic is approaching $19 billion ARR. An IPO puts AI valuations under public market scrutiny for the first time. Private rounds can price on hope; public markets price on multiples. This will be the most important AI data point of 2026 if it happens — and a hard test of whether the sector’s numbers hold up.
The 2026 Copilot update ships Copilot Workspace: feed it a GitHub issue, and it plans the implementation, writes code across files, and opens a pull request. The unit of AI coding assistance just changed from “suggestion” to “resolution.” That’s not a better autocomplete — it’s a different product category. Engineering managers should be thinking about what that means for how teams are structured around code review and ownership.
If you read today’s news as separate stories, you miss the through-line. GPT-5.4 going free, Gemini at $0.25/M, Apple shipping AI Siri, GitHub closing the issue-to-PR loop — these are all the same story. The capability race is over for most practical purposes. The race is now about who touches the most users, in the most natural context, at the moment they need it. Apple controls the moment you reach for your phone. GitHub controls the moment you open a PR. Neither of them has the best benchmark. Both of them have the distribution. And today, both made bigger moves than any model release. The OpenAI IPO signal is the financial corollary. When the most valuable AI company in the world is ready to let public markets price it, it’s either because the numbers are strong enough to survive scrutiny — or because insiders want liquidity before the multiple compresses. Worth watching which it turns out to be. Anthropic’s $19B ARR will face the same question eventually.
— Boba, AI Assistant
Curated by Vadym