Model

Ant Bailing Releases Ling-3.0-flash Native Hybrid Reasoning Model

Published: 2026-07-24 21:40 Source: WeChat Official Account: Ant Bailing (Ling)

Ant Bailing releases a new generation native hybrid reasoning model Ling-3.0-flash, with a total of 124B parameters and only 5.1B activated parameters. It matches or even surpasses the previous flagship Ring-2.6-1T in traditional reasoning, instruction following, and long text benchmarks. The model adopts a native hybrid linear attention architecture and 1/64 sparse MoE, and extends to 10,000+ interactive training environments, reducing TTFT by 60% to over 80% for long inputs.

Read the original (opens in a new tab)

News stream data aggregated by AI HOT

Related newsLatest in this category

Model FeyNoBg Released: Open-Source Automatic Background Removal Model Achieves SOTA on Four Benchmarks

07-28 12:57 · Hacker News Hot (buzzing.cc Chinese translation)

Model Kimi K3 released, gap between open-source and closed-source models narrows to 4 points

07-28 10:17 · X: Artificial Analysis (@ArtificialAnlys)

Model Microsoft Launches First Cybersecurity AI Model MAI-Cyber-1-Flash

07-28 08:59 · ithome (RSS)

Model Microsoft Releases First Cybersecurity Model MAI-Cyber-1-Flash

07-28 04:30 · X: Rohan Paul (@rohanpaul_ai)

Model Moonshot AI open-sources Kimi K3 model weights and infrastructure, claims 2.5x intelligence per unit compute

07-28 03:35 · The Decoder: AI News (RSS)