Skip to content

Kimi K2.7 Code

Modified MIT

Moonshot AI · 1000B · transformer-moe

2026-06-12262K context1000B params

Use Cases

chatcodereasoningvisiontools

Quantization Options

QuantBitsVRAMQualityStatus
Q2_Krec2359.0 GBModerate
Q4_K_M4604.0 GBGood

About this model

Kimi K2.7 Code is Moonshot AI's coding-focused specialization of the K2-series trillion-parameter MoE (32B active, 256K context), released June 12, 2026. It targets long-horizon software-engineering agents: Moonshot reports a 21.8% gain over K2.6 on its own Kimi Code Bench v2 with ~30% fewer thinking tokens, making agentic runs cheaper and faster — though at release all published scores were first-party, with no independent SWE-bench numbers yet. Like K2.6, it is not a consumer-local model: community 2-bit GGUFs run ~339 GB and 4-bit ~584 GB, requiring server-class memory. On Ollama it exists only as the `kimi-k2.7-code:cloud` remote-inference tag; truly local runs mean Unsloth GGUFs in llama.cpp on workstation or server hardware.