Impressum
Why I Built This
I built llm-bench.io because finding the best local LLM for my setup consumed an absurd amount of time. I was searching for models that would fit my AMD Radeon RX7900XTX — balancing VRAM constraints with the need for good token speed and output quality. I also needed sufficient context length (80k+) and compatibility with tools like OpenClaw.
The research process was brutal. Download a model via Ollama, wire it to OpenClaw, test a few use cases, discover the output quality wasn't good enough or the speed was too slow, clean it up, and repeat. Over and over. Weeks of testing.
I searched for a website that compared LLMs across different hardware setups. Something that could aggregate real-world community data so I could learn from others' experiences. I didn't find it — not for local LLMs anyway. So I decided to build it.
Here we are. llm-bench.io is still early stage, but I hope it saves you the weeks I spent. Run the benchmark on your hardware, compare your results with others, and find the right LLM for your use case.
Information pursuant to § 25 MedienG and § 5 ECG (Austria)
Media owner and operator:
llm-bench.io
Thomas Kaindl
Austria
Contact
Email: [email protected]
Purpose of the website
Community-driven benchmarking platform for local LLM inference. Help developers and researchers find the best open-weight models for their hardware, use cases, and performance requirements.
Liability for content
The content of this website has been created with great care. However, we cannot guarantee the accuracy, completeness, or timeliness of the content. As a service provider, we are responsible for our own content on these pages according to general law. However, we are not obligated to monitor transmitted or stored third-party information or to investigate circumstances that indicate illegal activity.
Dispute resolution
We are neither obligated nor willing to participate in dispute resolution proceedings before a consumer arbitration board.