Frontier AI models are a few months ahead of open-source laptop-sized models
Open AI models you can run on your phone are better than any AI from 2024.
Open AI models you can run on your laptop are better than the best AI from six months before they came out.
Laptops can run frontier AI from six months ago
An open-weights model surpasses a proprietary frontier model from 4 to 17 months earlier, depending on your hardware, on the Artificial Analysis Intelligence Index (AAII).
| Hardware | Open Model | Size | Released | AAII | Frontier Model | Frontier AAII | Frontier Released | |
|---|---|---|---|---|---|---|---|---|
| Data center | Kimi K3 | 2.8T A104B | July 2026 | 44 | Claude Opus 4.8 | 42 | May 2026 | |
| 128GB laptop | Qwen3.8 Flash Next | 180B A6B | Aug 2026 | 40 | GPT-5.4 | 39 | Mar 2026 | |
| 40GB laptop | Qwen3.8 27B | 27B | Aug 2026 | 34 | GPT-5.3 Codex | 33 | Feb 2026 | |
| 10GB phone | K2 Horizon 7B | 9B | Sep 2026 | 21 | Claude 4 Opus | 21 | May 2025 |
(as of 1 Oct)
For all of the coding and tool use tasks that you would have given to o3 or Opus 4, you can give them to a model that can run on your phone.
For all of the coding and tool use tasks that you would have given to GPT-5.5 or Opus 4.7, in the spring of 2026, you can give them to a comparable model that can run on your laptop at the end of summer 2026.
How have local models been treating you? Let me know what you think!