/benchmark-models
/benchmark-models — a cross-model benchmark for gstack’s own skills: run the same prompt/skill through Claude, GPT (via Codex CLI), and Gemini side by side and compare latency, tokens, cost, and (optionally) an LLM-judge quality score. Auth detected per provider; unavailable providers skip cleanly; output as table / JSON / markdown; --dry-run validates without spending. (Standalone CLI: gstack-model-benchmark.)
Meta-tooling, not an SDLC-lifecycle capability — it evaluates gstack’s skills across models rather than building the user’s software, so it carries no implements: edge. It is the model-evaluation cousin of pattern-cross-model-review. Distinct from gstack-benchmark, which measures a web app’s runtime performance.
See Also
- gstack-benchmark — the unrelated web-app performance benchmark.
- gstack-codex — the cross-model reviewer this benchmarks alongside.