Use inclusionai/ling-3.1-flash for the verified OpenRouter route. First-party hosted pricing remains pending. The code below is a documentation example, not a successful inference run by this site.
What is Ling 3.1 Flash?
It is an inclusionAI text model listed by OpenRouter. The checked route describes text input and text output, with a 262,144-token context limit and up to 32,768 completion tokens. These are route limits, not a recommendation to send the largest possible request.
Use a small prompt first. Then evaluate the actual job: summarization, extraction or tool selection. Compare outputs against a fixed set of examples that includes malformed input and ambiguous instructions. Preserve failures instead of averaging them away.
What does this route cost?
The current model record shows OpenRouter input $0.00 and output $0.00 per million tokens, checked 2026-10-03. First-party pricing is not verified. A displayed zero token rate is a dated observation, not unlimited access or permanent free service.
Inspect the exact endpoint metadata before production use, and confirm rate limits and account terms separately. The first recorded rate is a baseline. Later verified changes on the same route appear in price history; no older paid price is guessed.
How do I send a first API request?
Set OPENROUTER_API_KEY in your local shell, not in client-side JavaScript, a public repository or the URL. Send this small request using the OpenRouter API shape. Save the HTTP status and full response. Check for an error body and usage fields before treating the returned text as a valid result.
curl --fail-with-body https://openrouter.ai/api/v1/chat/completions \
-H "Authorization: Bearer $OPENROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "inclusionai/ling-3.1-flash",
"max_tokens": 512,
"messages": [{"role": "user", "content": "Return a JSON object with language and summary for: Hello world."}]
}'The prompt asks for JSON, but it does not activate a guaranteed JSON mode. Keep the raw response and validate the extracted assistant content against a schema of your own. Do not send private customer data in a first connectivity test.
Which output and tool checks matter?
The checked route lists tools and tool_choice, but does not list response_format. Test the specific tool schema you plan to use. Check argument names, types, required fields and unknown tool names; a listed capability is not proof that every request will work.
A useful extraction check contains valid text, missing information and instructions embedded inside the input. Define the expected fields beforehand. Record whether JSON parses, whether required fields are present and whether values match the source text. If the application retries invalid output, include all attempts in its budget.
What has actually been tested?
| Check | What to record | Our result |
|---|---|---|
| Request availability | Exact model ID, channel, HTTP status, error body and timestamp | API inference not run |
| Output quality | Original prompt, complete output and predeclared pass criteria | API inference not run |
| Cost | Usage fields, input/output/cache tokens and rates at request time | API inference not run |
| Latency and retries | Elapsed time and failures per run; multiple runs before reporting a median | API inference not run |
Public documentation and route metadata were inspected on 2026-10-03. We did not submit an authenticated prompt, measure inference latency or verify tool-call accuracy. The example is ready for a controlled test, but no completion or benchmark number is presented as our result.
Run repeated samples on the same workload before reporting latency or success rate. Keep a baseline model, fixed inputs and a written scoring rule. Report which route and date produced the result, because provider routing and token rates can change independently.
API questions
Which Ling 3.1 Flash model ID should I use?
The checked OpenRouter endpoint identifies inclusionai/ling-3.1-flash. Do not append :free or substitute another Ling model without checking that exact route.
Can I enforce JSON with response_format?
The checked route does not list response_format support. A prompt requesting JSON is not an enforced schema; parse and validate the response in your application.
Has LLM API Prices tested its speed?
No authenticated API inference was performed. Public provider metrics are not our latency measurements.