Artificial Analysis Launches Progress Index: Open Models Trail Proprietary by 2-8 Months, Prices Collapse Tenfold

Benchmarking firm Artificial Analysis has released the Progress Index v1, which uses item response theory (IRT) to estimate models' general capability and plot it against release dates, providing a unified yardstick for language model evolution since 2022, released alongside the Intelligence Index v5. Data from the index shows frontier model scores rising from 125 to 325 over four years; open-weights models closing the gap with closed models markedly faster; and output prices at equivalent capability falling by orders of magnitude within months.

Confirmed

• The Progress Index shows GPT-3.5 Turbo (November 2022) scored 125; by late 2024 OpenAI first broke into a higher score band (the source material doesn't give full subsequent numbers; the index now tops out at 325)
• Since January 2025, every time an open-weights model set a new top score, it did so 2-8 months after proprietary models first reached that level
• Output token prices within a capability band dropped several orders of magnitude in months: GPT-5.3 Codex, the first model to reach 300, was priced at $14 in February 2026; equivalent models later fell to $0.28 — roughly a 50x drop
• By country: between July 2023 and September 2025, Chinese and US models traded the open-weights lead seven times; since September 2025 China has led continuously. DeepSeek V4.1 Flash scored 324, beating the best US open-weights score and the current open-source high

Why it matters

• The Progress Index provides a unified, cross-year, cross-vendor capability scale, making "model progress speed" and "price deflation speed" quantitatively comparable for the first time
• The open/closed gap shrinking to 2-8 months means proprietary models' capability lead window is closing fast
• Same-tier prices dropping tens of times within months vividly illustrates how steep inference-cost deflation is — useful reference for app-layer pricing and compute demand forecasts
• Chinese open-source models' sustained lead since September 2025 is a key data point for tracking shifts in the US-China open-source race

2026-10-30 ~ 2026-10-30 · 6 related posts

• [source] Progress Index: Frontier AI Capability Climbed From 125 to 325 in Four Years — ArtificialAnlys · 2026-10-30
• [source] Same-Capability Output Prices Fell 50x in Months: $14 to $0.28 Per Million Tokens — ArtificialAnlys · 2026-10-30
• Open weights trail proprietary LLMs by 2-8 months as same-capability prices fall 50x — ArtificialAnlys · 2026-10-30
• [source] China has led open-weights LLMs since Sept 2025 as DeepSeek tops US best by 17 points — ArtificialAnlys · 2026-10-30
• China has led open-weights AI since Sept 2025 as DeepSeek V4.1 Flash 1017 scores 324 — ArtificialAnlys · 2026-10-30
• Artificial Analysis launches Progress Index, a single capability scale for every LLM since 2022 — scaling01 · 2026-10-30