Open weights trail proprietary LLMs by 2-8 months as same-capability prices fall 50x
ArtificialAnlys · x · 2026-10-30
Artificial Analysis shared two key observations from its Progress Index:
- Open-weights catch-up: Since January 2025, every new highest open-weights score has arrived 2-8 months after proprietary models first reached that level. The current open-weights leader is DeepSeek V4.1 Flash 1017 at 324, a level proprietary models first hit in June 2026.
- Price collapse: Output token prices within a given capability tier have fallen orders of magnitude within months — GPT-5.3 Codex, the first model above 300, charged $14/M output tokens in February 2026; by July 2026, DeepSeek V4 Flash 0731 offered the same tier at $0.28, a 50x gap.
Takeaway: the open-vs-proprietary capability gap is shrinking to months, and capability commoditizes at astonishing speed.
More from Venture
- Phonic launches speech-to-speech model Phonic-1, raises $27.5M Series A — rohanpaul_ai · 2026-10-30
- Cohere Targets CAD $7B Series E at CAD $35.7B Valuation, Canada's Largest Round — JeffDean · 2026-10-30
- Anthropic files for potential $2T IPO after $16.3B H1 revenue, first $1B operating profit — PeterDiamandis · 2026-10-30
- Unitree Q3 revenue up 69.9%, AgiBot discloses ~$300M H1 2026 revenue — Scobleizer · 2026-10-30
- Thrive's ADT buy praised as a smart, AI-forward move by Joshua Kushner — saranormous · 2026-10-30
- Cohere's investor list is mostly government-linked buyers, notes analyst — Afinetheorem · 2026-10-30