Kimi K2 Instruct

released 2025-07-12

BenchAtlas Index

as of 2026-07-14

Not enough benchmark coverage for a composite score (needs ≥4 families across ≥3 categories). The per-benchmark evidence below stands on its own — sparse coverage is an evidence gap, not a low score.

Benchmark evidence

2 results

external indices

ECI

140.4 points

independent

Epoch AI Benchmarking Hub

factuality

MASK

contamination=Potential contamination warning: This model was evaluated after the public release of MASK, allowing model builder access to the prompts and solutions.

46.7 %

45.1–48.3

independent

Scale Labs

Agent + model results

systems, not bare-model scores

agent + model CodeSweep - SWE-agent + Kimi K2 Instruct	SWE-bench Verified	53.4 %	unverified	SWE-bench Leaderboard
agent + model mini-SWE-agent + Kimi K2 Instruct	SWE-bench Verified	43.8 %	community	SWE-bench Leaderboard
agent + model mini-SWE-agent + Kimi K2 Instruct	SWE-bench bash-only	43.8 %	community	SWE-bench Leaderboard
agent + model mini-SWE-agent + Kimi K2 Instruct	SWE-bench Verified	0.5 usd_per_task	community	SWE-bench Leaderboard
agent + model mini-SWE-agent + Kimi K2 Instruct	SWE-bench bash-only	0.5 usd_per_task	community	SWE-bench Leaderboard

These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.

Related variants

Kimi K2 Instruct 0905