AZ Labs

Industry News

Ling 3.0 Flash Runs in Grok CLI Through OpenRouter

25 July 20265 min read

A practical model-routing experiment puts InclusionAI's Ling 3.0 Flash inside Grok CLI, with the request verified through OpenRouter and the same route working in Pi.

Listen
0:00 / 0:00
Grok Build model picker showing Ling 3.0 Flash Free in the AIMI project workspace
Grok Build model picker showing Ling 3.0 Flash Free in the AIMI project workspace

Key takeaways

  • check_circleLing 3.0 Flash can be used from Grok CLI through a configured OpenRouter route rather than a native xAI model entry.
  • check_circleThe working route is inclusionai/ling-3.0-flash:free, selected in Grok Build through the ling-3 selector.
  • check_circleAIMI makes the experiment repeatable by recording routes, pricing, evidence, health checks, and harness compatibility in one catalogue.

A good model in an unexpected place

Ling 3.0 Flash is a new model from InclusionAI that is designed for fast, agentic work. In our testing, it felt like a useful model to have around: quick enough for interactive work, capable enough to be interesting, and available through a free OpenRouter route at the time of testing.

The experiment was simple. We wanted to see whether a model that was not natively exposed by Grok Build could still be used from the Grok CLI without changing the workflow we already liked.

How Ling is running in Grok CLI

Grok Build exposes a Ling selector in its model list. In this setup, the selector is mapped to OpenRouter's exact model route: inclusionai/ling-3.0-flash:free. Grok then sends the request through OpenRouter using its OpenAI-compatible API.

The successful smoke test used the Grok CLI selector ling-3 and asked the model to return a short confirmation. It returned OK over the OpenRouter route in 5.9 seconds. The screenshot shows the model selected in Grok Build, while the test result confirms that the route responded successfully.

What Ling 3.0 Flash brings to the test

The model maker describes Ling 3.0 Flash as a hybrid-reasoning mixture-of-experts model with 124 billion total parameters and about 5.1 billion active parameters. The published model information lists a native 256K context window, with longer-context support described separately by the model team and hosting platforms.

Those numbers are useful context, but the practical test is simpler: does it respond quickly, follow the request, and behave reliably inside the tool you actually use? That is why the route and harness test matter more here than a model card alone.

Why the model index matters

AIMI is the catalogue behind the experiment. It keeps the model identity separate from the provider route, so Ling can be tracked across OpenRouter, OpenCode Zen, Pi, Grok CLI, and any other compatible harness without treating those routes as the same thing.

The index also records pricing semantics, endpoint first-seen times, evidence captures, health results, and harness-specific tests. That makes it much easier to answer a practical question: where does this model work right now, and what happened the last time it was tested?

Try the same setup

The exact Grok selector used in this test is ling-3. The route behind it is inclusionai/ling-3.0-flash:free on OpenRouter. You need a valid OpenRouter API key in the environment used by Grok Build, and the route may change because free provider availability is not permanent.

For a repeatable setup, keep the provider route in a model catalogue rather than hard-coding it in one tool. That gives you a place to record the route, verify it, and move it to another harness when the provider or model changes.

Frequently asked questions

Is Ling 3.0 Flash a native Grok model?

No. In this setup, Grok Build uses the ling-3 selector while the request is routed to InclusionAI's Ling 3.0 Flash route on OpenRouter.

Is the OpenRouter route permanently free?

Not necessarily. The route was verified as zero-priced during testing, but provider pricing and free availability can change. Check the live route before relying on it.

Where can I find the model catalogue?

The public AIMI repository is available at https://github.com/TH33ORACL3/aimi. The private catalogue also stores evidence captures, route health results, and local harness observations.

Sources

ShareShare on X