Start with a small, measurable workflow.
Useful evaluation cases include classifying incoming requests, extracting fields from documents, summarizing internal text and executing a bounded tool loop. Keep expected answers and failure criteria alongside each example. A valid JSON document can still contain incorrect facts.
This service uses an optimized derivative with additional quantization. Original-model benchmark scores do not establish the quality of this deployment, and Flonno does not publish a quality score for it.
Use the compatible API.
Send requests to /v1/chat/completions with a scoped Flonno API key and model ID qwen3.8-27b. Streaming and tool calls are supported. Native request details still matter; check the integration examples before replacing a different endpoint.
curl https://www.flonno.com/v1/chat/completions \
-H "Authorization: Bearer $FLONNO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3.8-27b",
"messages": [{"role": "user", "content": "Return a short JSON summary of this product."}],
"reasoning_effort": "none",
"max_tokens": 500,
"response_format": {"type": "json_object"}
}'Keep reasoning inside your budget.
Reasoning defaults to medium. Set reasoning_effort: "none" for a deliberate non-thinking request, or select a supported thinking setting for difficult work. Reasoning and visible output share the requested output budget; there is no separate thinking-token cap.
An omitted output allowance defaults to 4,096 tokens. Input and output share the 131,072-token context ceiling, so the maximum output is also limited by the input already present. Large contexts and concurrent requests can change observed latency.
Provide images inline.
Image inputs use image_url message parts containing PNG, JPEG or WebP base64 data URLs. The gateway accepts at most four images, up to 2 MiB per image, within its 4 MiB JSON request limit. Remote image URLs are not accepted. Image tokens use the input tariff.
For document extraction, test scans, small text and missing fields separately. Image understanding is an input capability; this API does not generate images.
Price the complete task.
At 10,000 requests with 1,000 input tokens and 300 billed output tokens each, the token estimate is $1.05: $0.30 input plus $0.75 output. Include billed reasoning, retries and rejected answers in your own estimate. This example excludes taxes and other workflow costs.
Flonno uses prepaid credit and reserves an allowance before generation. Actual usage and any request reconciliation determine the settled charge. A low token tariff is useful only when the output passes your acceptance criteria.
Calculate your own monthly token estimate →Review the processing path.
The Flonno gateway receives request content before routing it to the configured inference service. Model optimization does not establish confidentiality, TEE isolation or legal compliance. Review the current privacy notice and service terms before submitting personal or sensitive data.
These limits document the configured Flonno API contract, checked on October 10, 2026. They are not an uptime guarantee or a throughput promise.