Open Source AI Expansion and Pushback Against Big Labs Gains Momentum

오픈소스 | Sat Aug 15 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 5 sources

Meta released Glimmer, Chinese open models surged ahead, and Writer launched a new GLM-based model, expanding the open source AI ecosystem.

Analysis

[Meta] released open-weight model Glimmer ^[1]^[2]

Anyone can download and run it on their own hardware
The higher-tier Muse Spark model remains API-exclusive
Zuckerberg simultaneously published a 6
500-word manifesto

[Writer] launched Palmyra X6 based on GLM-5.2 with improved harness ^[3]

Built by post-training Z.ai's open source model GLM-5.2
Reduces baseline task costs by up to 50%
Harness optimization demonstrated an average 40% cost reduction

[Chinese Labs] led the frontier with massive open source models ^[4]

Released models ranging from 754B to 2.78 trillion parameters monthly
US remained under 130B for 5 out of 7 months
Moonshot
MiniMax
Xiaomi
and Z.ai rarely released models under 70B

[AMD & NVIDIA] led the US open source ecosystem ^[4]

Each released over 200 new model repositories
LiquidAI ranked third with around 100
Open model strategy aimed at driving hardware sales

[Hugging Face Hub] saw explosive growth in open model repositories ^[4]

Public model repositories grew from 2.43M to 2.96M
Datasets grew from 711K to 1M
Spaces from 1M to 1.44M
85.6% of models have fewer than 200 lifetime downloads

[Tim O'Reilly] criticized big lab control structures through open source AI ^[5]

Argued for opening the entire stack beyond open weights
Called for clear separation of model
harness
and application
Widespread societal diffusion of low-spec models is key to competitiveness

Sources