I track how much each provider charges for the same open LLM: Llama 3.3 70B ranges from $0.155 to $0.78 per 1M tokens

说明与用途

自动监控比较同一开源LLM在不同供应商的价格，突出最便宜的选项和量化差异。

问题

Different providers charge varying prices for the same open-weight LLM, making it hard to identify the cheapest option.

结果

The monitor identifies that DeepInfra offers Llama 3.3 70B at $0.155/M, the cheapest among 9 providers, with fp8 quantization on both ends, while other models show similar price disparities.

工作流

• Every 5 minutes read OpenRouter public per-model endpoint data (no API key).
• For each provider compute blended cost per 1M tokens assuming 3 input tokens per output token.
• Exclude endpoints that are down (negative status or 0% recent uptime).
• Compute median, mean, standard deviation per model.
• Flag providers 15% or more below the median.
• Warn if cheapest provider uses different quantization than most expensive one.
• Use n8n workflow with duplicate filter to avoid re-alerting the same provider/model pair for 6 hours unless price moves >5%.

证据

Every 5 minutes it reads OpenRouter's public per-model endpoint data (no API key needed).

来源

For each provider it computes a blended cost per 1M tokens, assuming 3 input tokens for every output token.

来源

It warns when the cheapest provider uses a different quantization than the most expensive one, because fp4 vs fp16 is not a fair comparison.

来源