Ling 3.0 Tiny
by inclusionAI · Model · Language & code models
Released free through an API with open weights promised soon; 7.9B total, 1.3B active parameters, MoE
Evidence
- Aug 6Hybrid reasoning model that switches between thinking and instant modes; 262K context; aimed at math, tool use and coding
- Aug 6Charted ahead of Qwen3.5 4B/9B and Gemma 4 on general agent, banking agent, hallucination and Olympiad math; behind on knowledge, science and long-context retrieval
- Aug 6Rejected four of five hallucination traps (fake paper, fake requests method, fake treaty, false Postgres claim); claimed an implausible April 2023 cutoff