GPT 5.4's performance on benchmarks

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> GPT-5.4 is OpenAI's first model with native computer use - can operate keyboard, mouse & applications directly on the user's behalf

> combines reasoning, coding & professional work (spreadsheets, docs, presentations); individual claims 33% less likely to be false vs GPT-5.2

> GPT-5.4 Thinking lets users tweak or redirect mid-response - no need to start over; rolling out to Plus, Team & Pro

> available now across ChatGPT, Codex, and API; GPT-5.4 Pro for max performance on Enterprise & Edu

> Anthropic now leads US business AI chat spend; Claude Enterprise, Team & Max all surging since mid-2025

> Claude Code writes ~4% of GitHub commits today (projected 20%+ by end-2026); Accenture deploying to 30,000 professionals + 8/10 largest US companies already on Claude

> however, Trump ordered all federal agencies off Anthropic due to latter refusing a DoD autonomous-weapons contract

> DoD officially named Anthropic a "supply chain risk", demanding all defense partners to cease use of Anthropic’s models

> threatens removal from MS Copilot & $60B+ in investor capital; Anthropic to sue in retaliation

> feud became marketing windfall for Claude: #1 App Store & Play Store, signups 4×, downloads +220% overnight

> ChatGPT uninstalls +300% the day OpenAI signed the Pentagon deal instead

> ceiling remains real: Claude holds ~4% of daily mobile users vs ChatGPT's ~42%, free users don't pay & no ads are planned

> according to SemiAnalysis, Anthropic's quarterly ARR additions already exceed OpenAI's, with full crossover projected mid-2026

> OpenAI's economics are tightening: margins fell 40% → 33% (vs. own 46% target), inference costs quadrupled, burn rises $8B → $25B → $57B through 2025-27

> OpenAI turns cash-flow positive only in 2030, currently raising $100B+ at $730B valuation; Anthropic gets there in 2028, constrained by compute supply, not demand

> Pentagon feud is the wildcard: could simultaneously hit Nvidia supply, Microsoft distribution & federal revenue just as Anthropic closes the gap - $60B in capital riding on the outcome