Model result · rank #10

Kimi K2.7 Code

OpenRouter · extra-high reasoning. Public result card with the model’s overall score, lane measurements, runtime/cost telemetry, and ranking formula.

Download source JSON Compare all models

Overall score73.28

Rank #10

Full / Agentic87.95

Full rank #4

SWE MVP58.61

SWE rank #11

Measured cost$0.569

100.0% reliability

Overall

All-around publication view

Score73.28

Formula50% Full + 50% SWE

BasisOpenRouter · extra-high reasoning

The overall score is calculated from the Full/Agentic and SWE lanes, keeping the aggregate comparable while preserving the measurements behind it.

Lane 01

Full / Agentic benchmark

Final87.95

Capability96.87

Agentic92.22

Pass rate97.7%

Prompts43

This lane captures instruction following, structured behavior, tool discipline, and general agentic reliability.

Lane 02

Software engineering MVP

SWE score58.61

Focused final57.66

Capability50.84

Daily driver59.24

Prompts24

This lane is closer to implementation usefulness: source handling, architecture cleanliness, and deliverable quality.

Telemetry

Runtime economics

Full cost$0.440

SWE cost$0.129

Full avg seconds12.71

SWE time860.76s

Decode25.23

Cost, time, and runtime basis are telemetry. They explain tradeoffs; they do not secretly overwrite the capability scores.

Interpretation

Why this result lands here.

The model is stronger in the Full/Agentic lane than in the SWE lane; the overall score is therefore shown with both component lanes visible. Exact OpenRouter model: moonshotai/kimi-k2.7-code. Full/Agentic lane is strong; SWE lane exposed patch-format and implementation-delivery failures in the MVP fixtures.