Access 24+ AI models through one OpenAI-compatible endpoint. Competitive pricing from a live marketplace, backed by multi-layer quality enforcement.
Why buy on Dragonfly
A spot market where price, reliability, and settlement are visible — not a black box with hidden margins.
Price
Sellers compete on a live order book, driving prices below official rates. You always pay the best available ask — no opaque markups.
Compatibility
Same OpenAI SDK, one base URL. Switch between DeepSeek, Qwen, GLM, Kimi, MiniMax and more with zero code changes.
Control
Bind cost, latency, and provider constraints to your API key. The matching engine enforces your rules on every request.
Visibility
Every request produces route_attempt, route_result, and settlement_result events. Debug cost and reliability from one ledger.
Six layers of protection between you and a bad response.
Every seller is scored on success rate, latency (p50/p90/p99), and availability in 30-minute windows. Anomalies are flagged automatically.
Relay responses are periodically compared against official provider output. We measure WER, CER, and cosine similarity per listing.
A percentage of every trade is held in reserve. If a seller underperforms, the reserve is consumed to compensate affected buyers.
Risk flags trigger auto-suspension or delisting. Sellers accumulate penalty points — unreliable capacity is removed from the book.
Choose quality_first routing and the engine weights health (35%), error rate (35%), and latency (30%) above cost.
Suspicious trades are flagged for review. Confirmed issues result in automatic buyer refunds from the seller's reserve.
From signup to first inference in under a minute.
Sign up, top up your balance, and generate a key. Bind a routing profile for cost, latency, or quality priority.
Change your base URL to Dragonfly. Same OpenAI SDK, same parameters. One line of code.
The matching engine finds the best eligible seller, executes the request, and settles the trade. You see every step.
Create a key, add $0.10 to start (or use the free trial credit), and make your first request. No minimums, no lock-in.