Sign in
Pending benchmarksTracking since Oct 27, 2026

Is Ling-3.1-flash-base state of the art?Pending benchmarks.Announced Oct 27, 2026; no independent results yet.

The base checkpoint of Ant Ling's Ling-3.1-flash: the pre-trained model before post-training, released as open weights under MIT for continued pre-training, domain adaptation and post-training research. A hybrid linear-attention mixture-of-experts model with 560B total and 25B active parameters. Not a chat model.

Announced facts

The details so far

Lab
Ant Ling (InclusionAI, Ant Group)
Announced
Oct 27, 2026
API
Not yet · None
Open weights
Published · Hugging Face inclusionAI/Ling-3.1-flash-base (BF16, 109 shards); announced Oct 27, 2026
Parameters
560B total / 25B active
Type
Base checkpoint (no post-training); BF16 weights after Warmup-Stable and Merge (WSM) checkpoint merging; trained in FP8
Architecture
Hybrid-linear MoE: 63 KDA + 9 Gated MLA layers (7:1), 4 dense layers, 512 routed experts with 8 routed + 1 shared active per token
Vocabulary
157,184 tokens
Relation
Base of Ling-3.1-flash and Ling-3.1-flash-1019 (tracked separately; their scores are not shared)
License
MIT

Tags marked unsure are not stated on the lab's own pages.

Next steps

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

  • Publicly announced with a source
  • Usable by the public (API or weights)
  • Listed on an independent board we track
  • Enough results for a rank in a category
  • Holds across two snapshots (crowns only)
Self-reported

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: evaluated on 47 benchmarks across eight domains against DeepSeek-V4-Pro-Base and MiMo-V2.5-Pro-Base, e.g. MMLU-Pro 76.71 (5-shot) vs 75.28 and 71.26.

Receipts

Sources

LABAnt Ling on X: Ling-3.1-flash-1019 open source, plus Ling-3.1-flash-basex.com/AntLingAGI/status/2115062200687878157 · Oct 27, 2026REPOinclusionAI/Ling-3.1-flash-base on Hugging Facehuggingface.co/inclusionAI/Ling-3.1-flash-base · Oct 27, 2026
Editorial

Compare when it lands

Unscored: the post-trained model, Ling-3.1-flash. A scored comparison appears once both are on a shared board.

Browse releasesRSSShareJSON

Tracking continues until an independent board lists it.