Chinese AI lab DeepSeek’s last model release was V3.2 (and V3.2 Speciale) last December. They just dropped the first of their hotly anticipated V4 series in the shape of two …
According to the article/blog they self-report lagging frontier models (ie. OpenAI and Anthropic) by 3-6 months. My personal experience has been that GPT 5.3 codex and later models are genuinely useful in daily programming, so that should mean a lot of options for good coding agents outside of Claude and codex in a few months.
In these times I can’t fault anyone for not having enough blood to sell to buy a GPU, but… IMO the utility of an open-weight model is that you can run it locally/privately
Available on Ollama cloud at https://ollama.com/library/deepseek-v4-pro .
According to the article/blog they self-report lagging frontier models (ie. OpenAI and Anthropic) by 3-6 months. My personal experience has been that GPT 5.3 codex and later models are genuinely useful in daily programming, so that should mean a lot of options for good coding agents outside of Claude and codex in a few months.
In these times I can’t fault anyone for not having enough blood to sell to buy a GPU, but… IMO the utility of an open-weight model is that you can run it locally/privately
I looked into running GLM 5.1 locally, but just the cooling system would probably cost more than a car, so I shelved it 😅