AdaL
    Available now in AdaL CLI and Desktop

    Code with Kimi K2.7
    in AdaL

    Moonshot AI's open-source coding-focused agentic model brings efficient long-horizon software engineering into AdaL. Use Kimi 2.7 for repo-aware implementation, refactoring, debugging, and verification.

    Get Started

    Run /model and pick Kimi 2.7 for coding tasks · 256K context · thinking-enabled reasoning

    View model docs Kimi K2.7 Code announcement
    Official Kimi K2.7 Code metrics

    Benchmarks and operating numbers

    Moonshot AI reports stronger coding scores, stronger agentic benchmark results, and lower reasoning-token usage compared with Kimi K2.6.

    62.0
    Kimi Code Bench v2

    +21.8% vs K2.6

    53.6
    Program Bench

    +11.0% vs K2.6

    35.1
    MLS Bench Lite

    +31.5% vs K2.6

    76.0
    MCP Atlas

    agentic benchmark

    81.1
    MCP Mark Verified

    agentic benchmark

    30%
    fewer thinking tokens

    average reduction vs K2.6

    256K
    context window

    262,144-token evaluation context

    $0.95 / $4.00
    API cache-miss input / output

    per 1M tokens, per Kimi API pricing

    Benchmark comparisons are reported by Moonshot AI against Kimi K2.6. K2.7 Code, K2.6, GPT-5.5, and Claude Opus 4.8 were evaluated in the tool settings documented in the official Kimi announcement.

    Why use Kimi 2.7 in AdaL

    A coding-specialized model for developers who need long-context repository work, efficient reasoning, and multi-step agent execution

    Purpose-built for agentic coding

    Moonshot AI positions Kimi K2.7 Code as an open-source coding-focused agentic model for complex, long-horizon software engineering tasks.

    1T-parameter MoE architecture

    K2.7 Code uses a Mixture-of-Experts architecture with 1 trillion total parameters and 32 billion activated parameters per token.

    256K context for large repos

    The model supports a 256K context window with Multi-head Latent Attention, making it a strong fit for repo-scale inspection and multi-file planning.

    Higher coding benchmark gains

    Moonshot reports a 21.8% improvement on Kimi Code Bench v2 and a 31.5% improvement on MLS Bench Lite compared with K2.6.

    More efficient reasoning

    K2.7 Code reduces thinking-token usage by about 30% versus K2.6, helping agent loops respond faster and lower operating cost.

    Vision-enabled coding context

    K2.7 Code incorporates MoonViT, a 400M-parameter vision encoder, so visual artifacts can support coding and debugging workflows.

    How to pick Kimi 2.7

    Available in AdaL CLI and Desktop through the model selector

    1

    Open AdaL CLI or AdaL Desktop

    Use the coding environment where AdaL can inspect your repo, run tools, edit files, and verify changes.

    2

    Run /model

    Open the model selector and choose Kimi 2.7 for coding-focused agentic workflows.

    3

    Start an engineering task

    Ask AdaL to inspect, plan, implement, test, and review with Kimi 2.7 powering the worker loop.

    What to know before you switch

    Selected details from Moonshot AI's Kimi K2.7 Code announcement

    "Kimi K2.7 Code is an open-source, coding-focused agentic model optimized for complex, long-horizon software engineering."

    Moonshot AI

    "Pick Kimi 2.7 when you want an efficient coding model for repo-aware implementation, refactoring, debugging, and verification loops."

    AdaL

    "The 256K context window is useful when AdaL needs to inspect larger code surfaces before making surgical edits."

    Context window

    "K2.7 Code always runs with thinking enabled and is purpose-built for coding; Moonshot recommends K2.6 for general writing and analysis."

    Important caveat

    Source: Moonshot AI Kimi K2.7 Code announcement.

    Open-source coding model for agent loops

    Ready to try Kimi 2.7 in AdaL?

    Open AdaL CLI or Desktop, run /model, and choose Kimi 2.7 when your task needs coding-focused long-context reasoning.

    Get Started with AdaL
    AdaL CLIAdaL Desktopofficial Kimi announcement linked