Ling 3.0 Tiny Supports Native Function Calling with Only 1.3B Active Parameters
Danare_113 · reddit · 2026-08-07
Ant's InclusionAI lab introduced Ling 3.0 Tiny, a model with 7.9B total parameters but only roughly 1.3B active parameters per token. Currently available only via API, it still supports native function calling at this extremely low activation budget, featuring a 256K context window and a 32K output cap. Additionally, it includes a switch between thinking and instant modes to reduce latency overhead when reasoning is unnecessary. The author highlights that the model size required for reliable tool calling is shrinking rapidly and questions where the actual floor lies.
Related event: AntLingAGI Releases Ling-3.0-tiny Native Hybrid Reasoning Model(2 posts)→
More from Models
- Cisco Releases Open-Weight Antares Models for Code Vulnerability Detection — aminkarbasi · 2026-08-07
- From Sycophantic to Condescending: Users Demand Straightforward AI Tools — bendee983 · 2026-08-07
- NVIDIA Launches Alpamayo 2 Super: A 34B Parameter Open Model for Robotaxis — emmanuelvivier · 2026-08-07
- Google Reportedly Preparing Another Flash Lightweight Model — Rare_Bunch4348 · 2026-08-07
- MiniMax H3 video generation costs under one-third of rivals, 15s 2K clip around $1 — eyishazyer · 2026-08-07
- Video generation model rankings: Seedance 2.0 leads, MiniMax H3 third — eyishazyer · 2026-08-07