LLM prices collapse 99.7% in 3 years: GPT-4-class capability now $0.10/M tokens
AlexTensor · x · 2026-09-21
AlexTensor argues AI uncertainty demands rigorous risk management, not VC-style bailouts disguised as regulatory "safety" policy: moats around closed-LLM firms and steering retirement savings into companies whose pricing power collapsed from $30 to $0.10/M tokens repeats past bubbles' moral hazards. Risk capital should absorb its own downside.
The linked analysis details the pricing collapse:
- GPT-4 launched at $30/$60 per 1M input/output tokens (Mar 2023); Gemini 3.1 Flash now costs $0.10/$0.40 — a 99.7% drop in three years.
- Frontier models too: Claude Sonnet 4.6, which beats GPT-4 on every benchmark GPT-4-era, costs $3/M input; Gemini 3.1 Pro $2/M — 93–97% cheaper than the priciest option of 2023.
- The constraint has shifted from "can we afford API calls" to "can we build infrastructure for cheap intelligence": multi-model routing, cascade verification, and speculative execution are now viable for modest startups, plus smart routing and caching strategies.
More from Venture
- Outrank Quietly Scrubs 'Backlink Exchange' Wording After Viral Warning — doooyle · 2026-09-21
- Indie dev adds supplier auto-detection and duplicate merging to finance app — kevinkern · 2026-09-21
- Open models hit 78.4% of token volume on Vercel AI Gateway; Chinese labs' spend tops OpenAI — burny_tech · 2026-09-21
- Sora's API shuts down as Kling raises $3B at $18B valuation: AI video's consumer flop, B2B boom — Expertiez0604 · 2026-09-21
- FAL Interview: MiniMax H3 Max Won't Be Open-Sourced, Going Frontier Commercial — pilkyton · 2026-09-21
- HVM creator Victor Taelin shuts PRs, flies to US to raise VC funding — TheMoonMidas · 2026-09-21