Damn, DeepSeek is down today. What can replace it?
Zhipu—the others work, but they can’t compete.
There doesn’t seem to be a complete replacement.
Let’s talk practicalities: there’s no complete replacement, but finding suitable alternatives for different scenarios is good enough. Here’s what I’ve learned:
Reasoning/coding: Zhipu’s reasoning models and the Qwen3 series are both quite capable. In complex reasoning, my experience is that they’re not far behind deepseek-reasoner, while for coding, models in the Qwen3-Coder line are arguably even more reliable;
Long texts/documents: Kimi’s long context is genuinely excellent. For tasks like reading papers and organizing materials, it’s considerably more useful than deepseek;
Cheap and high-volume use: deepseek is already the king of cost-effectiveness here. If you’re only comparing prices, no alternative beats it—you have to consider whether the specific task justifies the cost.
One more suggestion: add some abstraction at the API layer, and make BASE_URL + MODEL_NAME configurable. Use OpenAI-compatible interfaces throughout, so if one provider goes down today, you can switch to another tomorrow—it takes five minutes. And don’t rely on benchmark scores when choosing a model; running your own real prompts for comparison is the most reliable approach.
2 Likes
Wasn’t it already back to normal? Keep the music playing and keep dancing.