Ling-3.1-flash
#122 en Modèles de langage FrontierUnknown · v3.1 · flash · depuis 30. September 2026 · 2× · vu le 02 oct. 2026
Ling-3.1-flash is a mixture-of-experts language model announced by Ant Group's InclusionAI on September 30, 2026, with roughly 560 billion total parameters and about 25 billion activated per token. It is a hybrid reasoning model positioned for agentic workflows, coding, search, office software and specialist applications (including healthcare), with a targeted context window of up to 1 million tokens. At launch it is only available via a two-week free trial capped at a 256,000-token context; model weights, license and independent benchmarks have not yet been published — InclusionAI has only stated an intention to open-source the model later. All performance figures so far are vendor-reported only.
Fonctionnalités
| Key Benchmark (%) | 1,673 Elo on GDPVal-AA v2.1; 75.16 on FrontierSWE; 65.35 on HealthBench Professional (vendor-reported) |
| Context Window (Tokens) | Up to 1,000,000 tokens planned; currently capped at 262,144 (256K) tokens in the free trial |
| License | Not yet published; no license file, no weights, only an announced intention to open-source later (as of Oct 1, 2026) |
| Multimodality | Text only (input: text, output: text); no multimodality documented |
| Platform | Available via Vercel AI Gateway, NanoGPT, Command Code, and Ant's own Ling Studio (ling.tbox.cn); no standalone official product page with downloads |
| Price per 1M Tokens | Via NanoGPT: $0.075 input / $0.22 output (cache read $0.015); free via Vercel AI Gateway through Oct 13, 2026 |
| Release Date | September 30, 2026 (announced) |