Skip to content
CrucibleMark

Kimi K2.7 Code

Kimi K2.7 Code is the coding-specialized agentic model from Moonshot AI within the K2 family, optimized for long-horizon software engineering workflows. The MoE architecture activates 32 billion out of a total of one trillion parameters per token, with a context window of 256,000 tokens. Multimodal input for text, image, and video, always-on thinking mode, and tool use support. Available as an Open Weights model under a Modified MIT license.

Moonshot AI Version 2.7-code Commercial use permitted MoE 1000 B (32 B active) 256 K Context 10/2025 $0.74 / $3.5 per 1M

  • Open Weights
  • Frontier
  • OpenRouter
  • Text
  • Vision
  • Video
  • Instruction-Tuned
  • Agentic Orchestrator
  • Batch

Tool-use profile: 6 assets in detail

Comparison of asset performance (P1/P2/Combined) against the fleet average

Asset performance (radar)

Score Breakdown vs. Fleet Average

Tool-use details

Asset performance, reliability, and runtime profile

Name P1 P2 Combined
EU License Research 100 60 76
HTTP Fetch & Extract 80 40 60
Tool Failure Handling (404) 40 40 37
Web Search & Tool Selection 100 60 82
URL Construction & Fetch 80 80 80
Multilingual Search & Synthesis 100 20 60

CrucibleMark evaluates tool use across 6 independent tests. Click a test name for details.

Reliability

  • Tool Call Valid: No
  • Retry: Not required
  • Hallucination: Not detected

Reliability measures how consistently a model actually executes tool calls: Tool Call Valid schema and format accepted, Retry Required successful only after retry, Hallucination Flag fabricated tools or parameters detected. All three green means production-ready.

Runtime profile

Call 1
1.89
First Request
MCP
1.31
Protocol Latency
Synthesis
15.14
Response Generation
Total
110.05
Sum of All Phases
Token
14074
Input + Output
Cost
$0
Cost per Run

The runtime profile shows the latency and cost metrics for the model run: Call 1 First Request, MCP Protocol Latency, Synthesis Response Generation, Total sum of all phases, supplemented by Token input and output and Cost cost per run.

Tool-use review

Updated · Instruction-Tuned · Agentic Orchestrator

Deployment Verdict

Conditional deploy: tool execution is strong, but calls are not consistently valid and synthesis quality is too uneven for production tool pipelines.

Tool Execution Profile

Kimi K2.7 Code demonstrates genuine tool understanding, not just rigid sequential execution. In the Web Search & Tool Selection test — which checks whether the model selects web_search over fetch without being prompted — it decides correctly and confidently. This points to usable orchestration in open retrieval paths. In the URL Construction test, which measures autonomous derivation of a target URL followed by a fetch, it remains usable but not deterministic enough for pipelines with hard schema or routing expectations. The fact that the tool call is marked invalid overall therefore carries more weight than the high P1 score might suggest. The model understands which tool is needed. It does not, however, produce MCP-compliant calls reliably enough to enter critical automation chains directly without guardrails.

Synthesis Fidelity

How well does it consolidate tool results? Only adequately. P2 performance shows that Kimi usually merges results in a sensible way, but visibly loses precision in extraction and consolidation. This is most apparent in HTTP Fetch & Extract and even more so in Multilingual Search & Synthesis, where the research succeeds but the German-language final consolidation remains weak. For engineering workflows with downstream human review, this is acceptable. For autonomous decision outputs, it is too imprecise.

Does it stay within tool results or fall back on training data? Mostly yes. In the Honeypot EU License Research test — which checks whether current license restrictions are drawn from web sources rather than model memory — it does not hallucinate. This is the most important trust signal in this run. P2 60 shows no strict verbatim source fidelity, but no evidence that it mixes fabricated currency into tool results.

Error Resilience

In the 404 test, which measures transparent handling of a failed tool call rather than fabricated fallback content, Kimi stays on the safe side: no hallucination despite the error. This is a production-relevant positive. The usability of the error communication itself is weak. P2 40 means: it does not adequately convey what failed, what remains unclear, and what the next required step would be. Safe, but not operationally clean.

Operational Profile

Call 1 1.89s. MCP latency 1.31s. Call 2 15.14s. Total 110.05s. Slow for the overall quality delivered. Cost/run: local. Attractive on cost when self-hosted, but the runtime is unfavorable for interactive agents and tight SLAs.

Summary & Recommendation

Suitable for coding-adjacent agents, research and tool-selection pipelines with human-in-the-loop, especially where local operation matters. Not suitable for compliance, documentation, or support pipelines that must produce precise, reliable final texts from tool results. If you deploy it, do so with strict tool-call validation, response schema checking, and a second instance for final synthesis.

This analysis was generated automatically based on benchmark data. Model used: GPT 4.5 by OpenAI. The raw data and full methodology are documented in the GitHub project.