Kimi on Berges AI
The open-weights model from Moonshot AI, known for long-context reasoning. Pick it from the sidebar and it answers every turn.
Kimi is a family of open-weights large language models from Moonshot AI. The Kimi line is best known for long-context reasoning. Early versions pushed the practical context window past a million tokens, and the family has kept investing in that direction.
Berges AI serves the current Kimi model in its deep-thinking layer. Pick Kimi from the sidebar and it answers every turn, reasoning visibly before each reply. Your choice is remembered.
The weights are published openly, which means anyone can audit them, fine-tune them, or host them independently. That auditability is part of why we use it.
Kimi specs
What Kimi is good at
Long context
Kimi handles very large prompts without falling apart. Useful when you need the model to reason across a full document or a long thread.
Step-by-step reasoning
Multi-step problems where you need to keep state and follow a chain are a Kimi strength. That's exactly what the deep-thinking layer is for.
Multilingual
Kimi is unusually strong on non-English work, particularly Chinese-English, and holds up well across many other languages.
When Kimi is the right pick
- → Working with long documents or transcripts
- → Step-by-step reasoning across multiple turns
- → Translation and multilingual writing
- → Tasks where you need the model to "remember" a lot of context
Things people ask about Kimi
Can I use Kimi without Berges AI?
Yes. Moonshot publishes the weights openly. You can run them yourself or use any open-weights host. Berges AI is just one place to chat with Kimi, with privacy as a default and encryption at rest.
When should I pick Kimi?
When the input is long, when the problem has several dependent steps, or when a smaller model's answer feels shallow. Kimi is the largest model we run, so each reply costs more of your plan's units.
Is Kimi the same as ChatGPT?
No. Kimi is an open-weights model from Moonshot AI. ChatGPT runs OpenAI's closed models. Different labs, different weights, different design choices.