Kimi K3-256k (kimi.com)

Article summary

Kimi Code has introduced a new model, Kimi K3-256k, which has a context window of 256k and consumes about half as much quota as the full K3 model. The new model is suitable for everyday Q&A, code completion, and routine feature development. Users can switch between models, but switching invalidates the context cache and may require re-prefilling. The new model is available through Kimi Code and has different usage notes and recommendations.

Main themes

• Kimi K3-256k model
• Model switching and usage
• Kimi subscription waitlist
• Pricing and quota consumption
• Service quality and demand
• Comparison to other AI models and companies

What commenters say

• The new model is likely an API-level change rather than a retrained model.
• Some users are frustrated with the waitlist for Kimi subscriptions, but others appreciate the company's focus on serving existing customers.
• The demand for Kimi is high, and the waitlist is not a marketing gimmick.
• The new model's pricing is not publicly available, but it consumes about half as much quota as the full K3 model.
• Some users prefer Kimi's approach to limiting subscriptions to maintain service quality over other companies' practices.
• The new model may not offer significant advantages for users who already keep their context size below 256k.