localLLMriglocal, on your table

Compare

All the models are ready.

Pick any, run your prompt, see the answers side by side.

33 open models ready. Pick any, run your prompt, see the answers side by side.

9 of 9 open models selected. Nine open models is the most one comparison can run. Remove one to add another.

Small & fast
Mid-size
Large & smart
Code
Multilingual
Mixture of experts
Frontier reference

Do not paste personal or confidential data. This site does not store the prompt by default. Without a connected model provider the run is a labelled demo and the text is not sent out. When a provider is connected, the prompt is sent to OpenRouter and the pinned provider for this request and is not stored afterwards.

Worst case, list price, cap 2,000 in / 1,000 out: $0.017357€0.015504 · Claude Sonnet 5.5 $0.014€0.012506. EUR is an FX estimate.

Day Pass $9, 24h, max 50 comparisons. Month Pass $29, 30 days, max 300 comparisons.

Buy buttons stay hidden until checkout is available. The free mini-compare still runs.

Not sure? Tell us your use case and we'll suggest models

Optional. This picks nine open models from catalog tags. It does not score them.

Languages

Example prompts

You need a model tagged for reasoning, then the smallest one that is good enough.

  • Catalog label for this job: reasoning.
  • No budget was set, so size does not change the rank. The nine follow the use-case tag, languages, and context.
  • General versus specialist: this job is routed as reasoning. That is a routing label, not a benchmark.
  • English only. A multilingual catalog tag is not required for this request.
  • The requested context is one the shorter pinned endpoints can still cover. Each why line still states that endpoint's context.
  • An API is allowed. Local hardware is optional. A rented GPU is priced only from a checked hourly rate for a GPU that fully fits.
  • LFM2.5 2.6B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (65,536 tokens) covers the request.
  • Qwen3.5 9B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (262,144 tokens) covers the request.
  • Qwen3 14B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (40,960 tokens) covers the request.
  • Phi-4. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (16,384 tokens) covers the request.
  • gpt-oss-20b. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (131,072 tokens) covers the request. Mixture of experts: 21B total, 3.6B active. A memory estimate uses the total.
  • LFM2.5 24B A2B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (128,000 tokens) covers the request. Mixture of experts: 24B total, 2.45B active. A memory estimate uses the total.
  • GLM 4.7 Flash. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (200,000 tokens) covers the request. Mixture of experts: 30B total, 3B active. A memory estimate uses the total.
  • Nemotron 3 Nano 30B A3B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (262,144 tokens) covers the request. Mixture of experts: 30B total, 3B active. A memory estimate uses the total.
  • Qwen3 32B. Catalog tag: reasoning. Routing label, not a benchmark. Listed context (131,072 tokens) covers the request.

Built with Llama. Llama 3 Community License

Compare models · localLLMrig