Throttle blog https://www.throttle-pro.com/blog Measuring what self-hosted LLM inference really costs. en I changed nothing and my LLM server got 27% more expensive https://www.throttle-pro.com/blog/i-changed-nothing-27-percent-more-expensive https://www.throttle-pro.com/blog/i-changed-nothing-27-percent-more-expensive Fri, 25 Sep 2026 16:00:00 +0000 Four cost checks of the same untouched Ollama server came back at $8.77, $8.86, $11.26 and $10.02 per million output tokens. Why same-server runs swing, and the noise bound I now use before believing any before/after number. How to measure cost per million tokens on a self-hosted LLM (vLLM, SGLang, Ollama) https://www.throttle-pro.com/blog/cost-per-million-tokens-self-hosted-llm https://www.throttle-pro.com/blog/cost-per-million-tokens-self-hosted-llm Fri, 25 Sep 2026 16:00:00 +0000 A practical guide to turning GPU dollars per hour into dollars per million tokens for vLLM, SGLang and Ollama, the four pitfalls that make the number wrong, and how to re-check it after every config change and in CI. One vLLM flag, −68.6% cost per token: what a counterbalanced benchmark looks like https://www.throttle-pro.com/blog/one-vllm-flag-counterbalanced-benchmark https://www.throttle-pro.com/blog/one-vllm-flag-counterbalanced-benchmark Fri, 25 Sep 2026 16:00:00 +0000 A real A100 run where changing vLLM's max_num_seqs from 1 to 8 took cost from $0.746 to $0.234 per million output tokens, measured with a six-position B-C-B / C-B-C protocol. How the protocol works, the raw numbers, and what the result does not mean.