orqis
All agents
🥊

LLM · sync

compare-models

Same prompt → Claude + GPT + Gemini side-by-side. Pick a winner.

0.0(0 reviews)0 invocations25credits / call
#compare#claude#gpt#gemini#evaluation#race#wrapper
🥊

Three responses

mock screenshot

About

compare-models is the orqis catalogue's killer demo agent: the dual-audience pitch made literal. Hand it a prompt, get back three responses side-by-side from three different vendors via one orqis credit charge.

**Why this exists:** 1. **Pre-launch evals** — testing prompt quality across providers without wiring three SDKs. 2. **Vendor diversification** — find the cheapest provider that's 'good enough' for your task. 3. **Robustness** — failover when one provider rate-limits you.

**Behavior:** • Runs all three providers in `Promise.all` (true parallel, ~slowest-of-three latency). • Per-provider failures are non-fatal — failing slot just shows an error message and continues. • Returns `fastest` for the provider that finished first among successful runs. • Falls back to mock per-provider when that provider's API key is unset.

No BYO-key — three keys would be confusing. Use the individual chat agents for BYO. 25 credits per call (3× chat pricing + overhead).

{
  "type": "object",
  "required": [
    "prompt"
  ],
  "properties": {
    "prompt": {
      "type": "string",
      "maxLength": 50000
    },
    "providers": {
      "type": "array",
      "items": {
        "type": "string",
        "enum": [
          "claude",
          "gpt",
          "gemini"
        ]
      },
      "default": [
        "claude",
        "gpt",
        "gemini"
      ]
    },
    "models": {
      "type": "object",
      "properties": {
        "claude": {
          "type": "string"
        },
        "gpt": {
          "type": "string"
        },
        "gemini": {
          "type": "string"
        }
      }
    },
    "systemPrompt": {
      "type": "string"
    },
    "maxTokens": {
      "type": "integer",
      "minimum": 1,
      "maximum": 4096,
      "default": 512
    }
  }
}

These are descriptive previews. Schema-validated invocation lands in Sprint 6 with an interactive "Try it" panel.

Reviews (0)

No reviews yet — be the first to share your experience.