benchmark-models

Cross-model benchmarking runs identical prompts against Claude, GPT, and Gemini to compare latency, token usage, and cost. You select a prompt source, confirm providers, and run the benchmark to review the comparative results.

Yes, optional quality assessment is available via a built-in judge to surface best-of-show results. You must make an explicit choice during the interactive workflow to enable the judge before running the benchmark.

To benchmark multiple AI models, you need the gstack-model-benchmark binary and active provider logins for the models you want to evaluate. The interactive workflow will prompt you to confirm these providers before execution.

Cross-model evaluation for gstack skills supports prompting strategies, model selection, and optional quality scoring. You run a benchmark against a chosen gstack skill or inline prompt to evaluate and compare the different outputs.

Cross-model benchmarking requires the gstack-model-benchmark binary and explicit provider logins. You must also decide whether to enable the judge during runs, as quality scoring is optional and not enabled by default.