EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Jul 26·5 min read
Kimi K3 is not cheap
TL;DR
Kimi K3, a new open-weights large language model (LLM) from Moonshot AI, is not as cheap as some claim, costing more than other Chinese models. While it shows promise, particularly in coding tasks, its overall expense and performance nuances challenge the narrative of affordability.
✦ Why It Matters
Evaluate Kimi K3's capabilities and costs before considering it for your projects.
Key Takeaways
- Kimi K3 is not cheaper than US models despite claims.
- It costs more than other Chinese models like GLM-5.2.
- K3 is better for coding but more expensive for office tasks.
- Consider fine-tuning K3 for specific applications with additional resources.
Full Summary
Kimi K3, released by the Chinese lab Moonshot AI, is a powerful open-weights large language model (LLM) that has generated significant discussion regarding its cost and performance. Contrary to claims that it is cheaper than US models, K3 actually costs slightly less per task than OpenAI's top model and is significantly more expensive than other Chinese models, such as GLM-5.2 and DeepSeek V4.
While K3 excels in coding tasks, it is more costly for office-related tasks and tends to be verbose, leading to longer problem-solving times. Its open-weights nature allows for fine-tuning with additional GPUs, but this does not negate its overall expense.
The model represents a notable advancement for Chinese LLMs, raising concerns about its implications for the US market, especially given its rapid development with potentially lower capital expenditure.