The Lab Notes

InclusionAI

Ant's InclusionAI offers Ling-3.1-flash free through Oct. 13

Ant Group's InclusionAI is offering its 560 billion-parameter Ling-3.1-flash free for two weeks through gateways, with no weights or outside evals yet.

The Lab Notes ·

A flat illustration of a round-bottomed laboratory flask with an orange-to-green gradient, tilted over a pattern of small red, green and blue flasks on black beside a purple panel of white curved lines.

Developers can run Ant Group’s new Ling-3.1-flash for free through Oct. 13, but only through hosted gateways, because the model weights have not been published.

TechNode reported Sept. 30, 2026, that Ant Group’s InclusionAI unveiled Ling-3.1-flash, a hybrid reasoning mixture-of-experts language model with 560 billion total parameters and about 25 billion active per token. AI Weekly reported the same figures the same day. InclusionAI’s own announcement could not be reviewed, so the details here come from the outlets and distributors named.

The free window ends Oct. 13

Vercel’s changelog on Sept. 30 said it added Ling 3.1 Flash to its AI Gateway, “free to use through October 13, 2026.” After the promotion, Vercel said, the standard model ID, inclusionai/ling-3.1-flash, begins billing. The inclusionai/ling-3.1-flash-free variant stops serving rather than charging.

OpenRouter’s model list, read Oct. 3, shows inclusionAI: Ling 3.1 Flash, created Oct. 2, at $0 per million input and output tokens. It lists a 262,144-token context, text in and text out, tool calling, reasoning and up to 32,768 completion tokens.

The long context sits behind a paid tier

TechNode, citing the Chinese outlet IT Home, reported that the design supports a context window of up to 1 million tokens, while the two-week free trial is capped at 256,000. Vercel lists a 262K-token window. AI Weekly wrote Sept. 30 that the 1 million token window sits behind a paid tier launching after Oct. 13.

TechNode said InclusionAI intends to expand the window and release the model as open source after the trial. That release is a stated plan. AI Weekly said the weights were unpublished as of its article.

Uses and the benchmark record

TechNode said InclusionAI is positioning the model for agent tasks, search, office software and specialist applications. Vercel describes coding, multi-step analysis and agents that use tools, including workflows over long documents, code and extended task histories, and said it can be used from coding agents such as Claude Code and Codex.

AI Weekly wrote that all 8 publicly visible benchmark scores for the model are vendor-reported, taken from Ant’s own launch screenshots, and that no third-party evaluation has occurred. It quoted OrcaRouter calling Ling-3.1-flash “an announced one” rather than an open release, with no weights and no external evals. OrcaRouter compared it to Ling-3.0-Flash, which took two weeks to reach open source.

Analysis

We think the trial suits testing more than building. A mixture-of-experts model splits its parameters among many specialist sub-networks and uses only some of them for each token, which is what “25 billion active” means. Vercel said billing starts after Oct. 13 on the standard model ID while the free ID stops serving, so anything built on the free ID has to move that day. InclusionAI has tied the open-source release to the end of the trial, per TechNode, which makes it a plan until the weights appear. AI Weekly said all 8 benchmark scores are Ant’s own, so we would test it on a developer’s own tasks before relying on them.