The AI models with the lowest API input price across every lab Vikshy tracks, cheapest first. 4 offer a rate-limited free tier, and the lowest per-token paid price is Qwen3.7 Flash (Alibaba (Qwen)) at $0.03 per million input tokens. Ranked purely by listed input price; open-weight models (which you self-host) are covered in open-source models.
The lowest-cost options are free tiers (rate-limited): GLM-4.7-Flash (Zhipu AI (GLM)), GLM-4.6V-Flash (Zhipu AI (GLM)), GLM-4.5-Flash (Zhipu AI (GLM)). Among models with a per-token price, Qwen3.7 Flash (Alibaba (Qwen)) is the lowest at $0.03 per million input tokens. Cheaper is not better for every task; check the context window and what the model is built for.
Are open-weight models cheaper?
Open-weight models have no per-token API price because you run them yourself, so hosting cost replaces API cost. See the open-source models list.