Work in progress: Agents Directory has just launched. Stay tuned, more content is on the way.
Kimi K2.7 Code
Moonshot's coding-specialized K2.7, a 1T-parameter open-weight MoE (32B active) built for agentic tool use under a modified MIT license.
Kimi K2.7 Code launched on June 12, 2026 and is live on OpenRouter (hosted by Moonshot AI and Novita). Coding leaderboards like CursorBench, FrontierCode, and Artificial Analysis have not scored it yet, but the first independent signal is in: on ulam.ai's ErdosBench math-proof smoke test (14 problems) it placed 2nd of seven, behind Claude Fable 5 and ahead of GPT-5.5 (xhigh), Qwen 3.7 Max, Kimi K2.6, and Claude Opus 4.8. Moonshot's own Kimi Code Bench v2 puts it at 62.0, up from 50.9 for Kimi K2.6, with Opus 4.8 at 67.4 and GPT-5.5 at 69.0. Moonshot has also shipped Kimi K2.7 Code HighSpeed, a 6x faster serving mode (around 180 tokens per second on coding tasks) now in limited beta.
Moonshot's June 12 launch chart. Starred (*) rows are Moonshot's own internal benchmarks; GPT-5.5 and Opus 4.8 are shown at their xhigh reasoning settings. These are vendor numbers, not yet independently reproduced.
Source: Moonshot AI
OpenRouter
$0.74 / $3.5 per MAAkashml
$0.75 / $3.5 per MAAmbient
$0.84 / $3.99 per MAAtlascloud
$0.95 / $4 per MCCloudflare
$0.95 / $4 per MDDecart
$0.68 / $3.41 per MDeepInfra
$0.74 / $3.5 per MIInceptron
$0.75 / $3.5 per MIIo Net
$0.95 / $4 per MMMoonshot Ai
$0.95 / $4 per MNNovita
$0.95 / $4 per MPParasail
$0.75 / $3.5 per MMoonshot API
$0.95 / $4 per MSSiliconflow
$0.94 / $4 per MTogether AI
$0.95 / $4 per MVVenice
$0.9 / $4.3 per M
Provider
Moonshot AIContext window
262KInput price
$0.74/MOutput price
$3.5/MOpen weights
YesReleased
Jun 2026
