Open-weight model prices fall fastest at the cheapest endpoints
The cheapest output prices for open-weight AI models fell 17% in 30 days, while the median declined 1%, according to American Compute’s Open-Weight LLM Price Index. The changes are weighted by each model’s token usage and cover models listed throughout the period, highlighting a sharper decline at the low-cost end than in typical pricing.
Provider selection can make a substantial difference even for identical model weights. DeepSeek V4.1 Flash 1017, the index’s most-used model, had 40 providers listing output prices from $0.15 to $1.20 per million tokens, around a median of $0.71. Price movements also varied by model: Nemotron 3.5 Super VL recorded a median output price change of -20%, while many indexed models showed no change.
The index tracks daily list prices through a large public LLM router, net of listed discounts, using each provider’s cheapest paid endpoint at the end of each UTC day. It includes models with at least five paid providers, with prices current as of November 8, 2026. Direct and enterprise pricing are excluded, limiting the findings to that distribution channel.