theo counters Zeeg: same task 55x cheaper in a year with Gemini 3 Flash vs o1
theo · x · 2026-10-08
Zeeg argued intelligence hasn't gotten orders of magnitude cheaper, comparing models over 3 years. theo pushes back: 3 years ago there were no reasoning models or tool calls, so that scale is meaningless. He ran the same "name skateboard tricks" task on o1 (Dec 2024, $15/M in, $60/M out, cached reads $7.50) at $0.109 per question; Gemini 3 Flash (Dec 2025) scored higher at $0.0020 — 55x cheaper in a year. He suggests trying real code work on GLM-5.3-flash, which is 10x+ cheaper still.
Related event: theo argues same-level AI intelligence got 55x cheaper in a year(2 posts)→
More from Models
- Perplexity releases open multimodal late-interaction embedding family with SOTA results — antoine_chaffin · 2026-10-08
- Perplexity ships open-weights Decider v1.1: top Decision Index score, half the price — gaganghotra_ · 2026-10-08
- User forced to clarify 'we don't have that kind of relationship' with the model — repligate · 2026-10-08
- Closed frontier models lead open-weights by 3-6 months, agentic timeline shows — JosephJacks_ · 2026-10-08
- GPT-6 rolls out to Plus/Pro/Enterprise today, free users start tomorrow — michpokrass · 2026-10-08
- GPT-6's interleaved thinking answers instantly while it keeps reasoning — michpokrass · 2026-10-08