Breaking

Moonshot AI Releases Kimi-K2.7-Code, Ranking 3rd Among Open Models on Code Arena Frontend

June 15, 2026 at 16:19 EDT

Foundation Models
Software Dev & Coding
Open Source

Around June 12, 2026, Moonshot AI released Kimi-K2.7-Code, an open-weight model specialized for coding, which landed at 3rd among open models and 19th overall on the community evaluation platform Code Arena: Frontend. Code Arena: Frontend evaluates models on agentic frontend coding tasks—real users building apps and websites in HTML and React—reflecting real-world feedback rather than benchmark scores alone.

June 12, 2026 · Moonshot AI

Kimi-K2.7-Code: an open-weight coding model breaks into the top ranks

Moonshot AI's coding-focused, agentic model lands at No.3 among open-source models (No.19 overall) on Code Arena: Frontend — while cutting "thinking" tokens ~30% versus its predecessor.

Among open-source models on Code Arena: Frontend (No.19 overall)

−30%

Thinking-token usage vs K2.6 — less "overthinking"

Total MoE parameters — 32B activated per token, 256K context

Benchmark gains: K2.6 → K2.7

Published scores across five internal coding benchmarks (higher is better)

50.9

62.0

Kimi Code Bench v2

48.3

53.6

Program Bench

26.7

35.1

MLS Bench Lite

72.8

81.1

MCP Mark Verified

69.4

76.0

MCP Atlas

K2.6 K2.7

Pricing per million tokens

$0.19

Input

$0.95

Output

$4.00

Thinking

HighSpeed mode (~180–260 tokens/s) slated to arrive soon. Open weights under a Modified MIT license; distributed via Hugging Face, Kimi API, OpenRouter, Fireworks AI, Ollama, and Cloudflare Workers AI.

What developers praise

Less overthinking, stable over long coding loops
High token efficiency — practical and cost-advantaged
Pairs easily with Claude Code, VS Code (Cline, RooCode)
Multimodal: generates code from screenshots; approaches Claude-class in frontend tests

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.