Loading the catalogue…
Google DeepMind's Flash-tier multimodal model with a 1M-token context window and always-on, effort-adjustable reasoning for text, image, video, audio, and file inputs.
Same model, one row per endpoint, your call. Sorted by Sovereignty Ladder rung, highest first. The default endpoint is whatmodelalone resolves to.
| Rung | Provider | Input / M | Output / M | TTFT | Tok/s | Reliability | Status |
|---|---|---|---|---|---|---|---|
| L0 | GoogleDefault | €1.29 | €7.75 | 2.14 s | 153 | 100.0% | Active |
Pick this model first.
Works — usually a fine pick.
Other models will serve you better.
— None mapped yet.
Fit assertions set by the curator pipeline. 45 use cases mapped.
Model catalogue
Stav speaks the same chat-completions API you already use. Set STAV_API_KEY and the curl below works. SDKs (openai-python, langchain, litellm) all work with the same base URL.
curl https://api.stav.ai/v1/chat/completions \
-H "Authorization: Bearer $STAV_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemini-3.5-flash",
"messages": [{"role": "user", "content": "Hello!"}]
}'Prefer the Anthropic SDK contract? The same model is also reachable at POST /v1/messages — Stav's gateway translates between Anthropic-shape and the underlying provider for non-Claude models.
Sovereign-EU inference by default; the Smart Router lifecycle penalty drains traffic away from deprecating endpoints automatically. Add X-Stav-App-Id: <uuid> to attribute requests to a registered app.
Inference runs in US on google. The model was created in US as well. Both training and inference fall under US compelled-disclosure regimes (CLOUD Act / FISA 702). Not appropriate for EU regulated-sector workloads without explicit legal sign-off.
Inference and (typically) training both happen under US jurisdiction — both CLOUD Act and FISA Section 702 reach. EU regulated-sector customers cannot use this combination for sensitive data without a derogation / case-by-case legal review.