Gemini 4 Neon Tops FutureSim by Spawning 500+ Subagents and 16K Tool Calls
maksym_andr · x · 2026-11-21
A quoted post reports Google's Gemini 4 Neon achieving the top score on the FutureSim benchmark by launching 500+ subagents, making over 16,000 tool calls, and writing simulations for 50+ questions spanning elections, sports, job displacement and markets. The poster pairs it with a "Google apology form" meme, poking fun at the brute-force subagent approach — the absurd agent count is half the joke.
More from Fun
- Two weeks into the new job, he built a tiny office cafe next to an 'AI Readiness' whiteboard — armano · 2026-11-21
- AI assistant Dot finds a repair guy, negotiates the price and books an appointment — VraserX · 2026-11-21
- Security researcher jokes: 'all security vulnerabilities will be suspended for Thanksgiving' — emax · 2026-11-21
- If LLMs are just stochastic parrots, adopting a real parrot might be cheaper than tokens — mjdramstead · 2026-11-21
- LinkedIn's AI slop problem is so bad, capping users at 10 comments a day looks like the fix — gaganghotra_ · 2026-11-21
- mitsuhiko nudges GitHub to add an AI agent to 4-year-old merge=union feature request — mitsuhiko · 2026-11-21