Clef-flash

Cloudflare's 9B open-weight decision model — latency-tuned, answers in tens of ms.

At a glance

• Provider: Cloudflare (open-weight)
• Question types: choice, score, noul
• Images: yes (up to 4)
• Context: ~64k tokens
• Backbone: Qwen3.5-9B (frozen) + vision encoder
• Price: $0.25 per 1M input tokens · output free

Unified API — one shape, switch with model

The simplest way to call Clef-flash: the unified endpoint. Same request and response schema across every model, with the provider's untouched reply included as provider_data.

curl -X POST https://jevtypesafeai.com/api/v1/decisions \
  -H "Authorization: Bearer jv_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "clef-flash",
    "state": "Order R-208 was charged twice. Please refund the extra payment.",
    "questions": {
      "route": { "type": "choice", "instructions": "Which team should handle this?",
                 "criteria": { "billing": "charges and refunds", "account": "login", "other": "anything else" } },
      "urgent": { "type": "noul", "instructions": "Is this urgent?" }
    }
  }'

Native API — Clef-flash's own request & response

Call Clef-flash directly at /api/v1/clef-flash and get its native response back, byte-for-byte. Billing is reported in X-Decision-* response headers.

curl -X POST https://jevtypesafeai.com/api/v1/clef-flash \
  -H "Authorization: Bearer jv_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "state": "Order R-208 was charged twice. Please refund the extra payment.",
    "questions": {
      "route": { "type": "choice", "instructions": "Which team should handle this?",
                 "criteria": { "billing": "charges and refunds", "account": "login", "other": "anything else" } },
      "urgent": { "type": "noul", "instructions": "Is this urgent?" }
    }
  }'

One jv_live_ key and one prepaid balance work across every model. Prices are what we bill per 1M input tokens; output is free. Clef-flash and its marks belong to Cloudflare.