Source
· Daily Brief
AI Daily Brief — 19 January 2026
A heavyweight Monday from China and IBM. Z.ai (Zhipu) ships GLM-4.7-Flash — a 30B / A3B MoE model (~3B active params per token) as the free-tier counterpart to the GLM-4.7 flagship, under the MIT licence. Headline benchmark: 59.2% on SWE-Bench Verified — more than doubling Qwen3-30B-A3B’s 22.0%. Designed to run on a single 24 GB GPU at 60-80 tok/s. IBM launches Enterprise Advantage — an asset-based consulting service to help organisations build, govern and scale internal agentic-AI platforms across AWS, Google Cloud, Azure, watsonx and both open and closed models.
Top stories
- Z.ai — GLM-4.7-Flash. 30B / A3B MoE · MIT licence · 59.2% SWE-Bench Verified (vs Qwen3-30B-A3B 22.0%) · single 24 GB GPU at 60-80 tok/s. via NYU Shanghai RITS
- IBM — Enterprise Advantage. Multi-cloud, multi-model agentic-AI consulting service across AWS, Google Cloud, Azure and watsonx. via IBM
- Veo 3.1 — production reviews. PetaPixel’s deeper look highlights native vertical 9:16 output and 4K at 60 fps. via PetaPixel
Compiled by AI Feed’s editor from publicly reported announcements on 19 January 2026.