Latest open artifacts (#23): Laguna S2.1, Inkling, & Kimi K3 show the utility of open models on the Pareto frontier
Interconnects · Florian Brand · 2026-08-02
Florian Brand's Interconnects newsletter surveys recent open-weight model releases—including Thinking Machines' Inkling 975B MoE, Moonshot AI's Kimi K3, Poolside's Laguna S2.1, and DeepSeek-V4-Flash-0731—and argues open model adoption is accelerating rather than consolidating as many had predicted.
Appears in
Extraction
Topics: open-weight-modelsmodel-releaseschinese-ai-labsai-ecosystem
Claims
- Predicted AI industry consolidation has failed to materialize; more companies than ever are training and releasing powerful open-weight models at scale.
- Thinking Machines' open model finetuning service generates hundreds of millions in annual revenue, demonstrating the commercial viability of an open-model business strategy.
- Kimi K3's noncommercial license requiring commercial agreements with inference providers could create regulatory leverage for future US government action against entities working with Chinese AI companies.
- DeepSeek released V4-Flash-0731 just one day after OpenAI cut its smallest model prices by 80%, beating competitors at the efficiency-performance frontier.
- Meituan's LongCat-2.0, trained entirely on Ascend 910 accelerators, is the first non-Huawei non-toy model trained exclusively on Chinese hardware.
Key quotes
The demand for tokens is incredibly high, and likely to increase as models get more efficient and unlock more possible use cases.
Having predicted consolidation for a long time, it now seems like a safer bet is to predict continued adoption, and try to imagine the role that open models play there.
Now their open model finetuning service is making hundreds of millions in revenue per year and they're releasing the best open-weight models built in the U.S.A.