Kimi K3-256k Model Now Available via Kimi Code
Original: Kimi K3-256k
Why This Matters
A more quota-efficient flagship model variant broadens accessibility for developers on standard Kimi Code membership tiers.
Moonshot AI's Kimi Code platform has launched the k3-256k model ID, a 256K context version of its flagship Kimi K3 model (2.8T parameters). It uses approximately half the quota of the full 1M context k3 model, and is available to all Moderato-tier members and above.
Kimi Code has introduced the k3-256k model, a 256K context variant of its Kimi K3 flagship coding model featuring 2.8 trillion parameters. According to the official documentation, k3-256k consumes roughly half the quota of the standard k3 (1M context) model, making it a cost-efficient option for everyday tasks such as Q&A, code completion, routine feature development, and single-file or small-file edits. Unlike k3, video input is not supported in k3-256k. The model is available to all Moderato-tier and higher members. The platform currently offers four model IDs: k3, k3-256k, kimi-for-coding (Kimi K2.7 Code), and kimi-for-coding-highspeed. Users switching from k3 (1M) to k3-256k are advised to manually run a compact operation first if the session context exceeds 256K, in order to preserve task context and avoid data loss. Kimi Code also notes that after any model switch, cached context from the previous model will not carry over, causing temporarily higher quota usage; starting a new session is recommended for optimal performance and lower consumption.