Token maths is still not mathing; new model releases and escapes
Plus, an open policy problem
Deepnote's Substack
Deepnote's Substack
Subscribe:.rss.atom.json.md.m3u.pls
Live Last read · last published · next check
Latest episodes
Claude’s inner dialogue, price wars at the frontier, and a broken benchmark
It was a great fortnight in tech for everyone, apart from people facing lawsuits over corporate espionage and a developer in South Korea who received a $16.6 million invoice from Anthropic (that’s one way to hit revenue targets!).
Watermelon Sugar, Jalapeños, and Fable-ous return
Sonnet 5 landed as the new default: Opus-level agentic vibes at a lower sticker price, yet somehow pricier per task in independent tests, slower than 4.6 under all the extra thinking, and roasted on Reddit and HN within hours as chattier, more adversarial, and a flat downgrade for a lot of people.
Anthropic's models get pulled offline, SpaceX becomes the backbone of frontier AI, and Satya Nadella says most companies are building on sand
This issue lands at an unusual moment.
I/O drops Gemini Omni + 3.5 Flash, Karpathy joins napkin-profitable Anthropic, & everyone gets an FDE thanks to Wall Street
Hi!
Anthropic's $100M cyber defense coalition, GPT-5.5, and the agent stack converging
Anthropic revealed a model it’s keeping locked away precisely because it’s too dangerous to release commercially and built a $100M industry coalition around it instead.
OpenAI and Anthropic eye 2026 IPOs, Cursor's Composer 2 beats GPT-5.4 on cost, and SlopCodeBench delivers bad news
What a leaky fortnight.
Meta delays Avocado frontier model to May, Cursor, Ramp, and Anthropic race to build agent infrastructure, and GPT-5.4 merges the model stack
OpenAI collapsed its fragmented model lineup into GPT-5.4, the first model to combine reasoning, coding, and computer use while simultaneously shipping Codex-Spark at 1,000+ tokens per second on Cerebras hardware, a quiet signal that inference infrastructure is diversifying beyond NVIDIA.
Grok 4.2’s four-agent architecture, Anthropic’s Programmatic Tool Calling, and OpenAI’s $110B raise reshape the agent stack
This fortnight, two scenes captured another local maximum of AI hype: YC partners in crab suits, and Sam Altman counting calories to compare training humans vs training LLMs (the internet promptly fact-checked it and did not award extra credit).
Opus 4.6 vs GPT-5.3-Codex, Meta Avocado pre-training tops open models, MiniMax $0.30/hr
Two frontier labs launched their best models simultaneously, and for many developers, something shifted.