Ant Group launches Ling‑3.1‑flash, a 560 billion parameter model, in free trial
Ant's InclusionAI says Ling-3.1-flash activates 25 billion of 560 billion parameters and will be open-sourced after a two-week trial, TechNode reported.
Ant Group's InclusionAI unit launched Ling-3.1-flash on Wednesday, TechNode reported. The model has 560 billion total parameters, with about 25 billion activated per token, according to the developer. InclusionAI pitches it for agent tasks, search, office software and specialist applications.
InclusionAI designed the model for a context window of up to one million tokens, but the trial caps it at 256,000 tokens, TechNode said. The trial lasts two weeks and costs nothing. The report gives no pricing for afterwards.
InclusionAI says it plans to widen the context window and open-source the model once the trial ends, according to TechNode. Nobody has confirmed a checkpoint or licence, so the open-weights claim is a plan rather than a release.
Commentator kimmonismus called it a solid open-source release from China on X, saying it comes close to GPT-5.6 Sol and Opus 5 on key benchmarks. That post says 500 billion parameters, which differs from the 560 billion in TechNode. The benchmark figures are unverified.
The same commentator noted the comparison leaves out newer models, including Opus 5.5 and GPT-6.1 Sol. No independent test results exist yet, and the benchmark claims may shift once evaluators report.