Ollama LLM Throughput Benchmark

Intro

LLM Benchmark is an open source benchmarking site that measures throughput performance of local large language models via Ollama. The project collects and presents inference throughput (tokens per second) for models run on desktop platforms. Results are organized by platform and hardware so you can see how configurations compare without marketing language.

Key Features

• Throughput-focused benchmarks reported as tokens per second, enabling direct comparison of inference speed across configurations.
• Cross-platform coverage with results pages for macOS, Linux, and Windows, and a dedicated macOS results page that supports filtering by model.
• Publicly visible example metrics and top performers; for example, an Apple M1 Max is listed at 134.71 tok/s for model gpt-oss:20b in the reported results.
• Open source framing with site navigation referencing GitHub and PyPI, indicating source and distribution channels are surfaced.

Who this is for

This resource is for developers, researchers, and system evaluators who need a practical way to measure or compare throughput/inference performance of local LLMs across different operating systems and hardware. It focuses on throughput metrics rather than broader evaluation dimensions, and reported outcomes are tied to the specific hardware and model configurations that were tested. Note that the site is tagged with pricing and privacy, but the website does not include explicit pricing details or a full privacy policy, and benchmark methodology details are not provided in the available content.