Skip to content

Benchmark

MCPMark Verified Filesystem Pass@1

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.

Models measured
8
Scores recorded
8
Sources
1
IQ weight
1.18

Score distribution

83.8%
Mean
86.7%
Median
7.7%
Std dev
76.7%
10th percentile
90%
90th percentile
70%
Min
70%93.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.18 for chat models, measured over 8 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1Kimi K3config: kimi-k3-max, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs93.3%
2Claude Fable 5config: claude-fable-5-max, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs90%
3GPT-5.6 Solconfig: gpt-5-6-sol-max, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs90%
4Claude Opus 4.8config: claude-opus-4-8-max, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs86.7%
5GPT 5.5config: gpt-5-5-xhigh, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs86.7%
6DeepSeek V4 Proconfig: deepseek-v4-pro-max, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs76.7%
7Kimi K2.7 Codeconfig: kimi-k2-7-code, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs76.7%
8Kimi K2.6config: kimi-k2-6, task-set: standard, task-version: verified, scope: filesystem· MCPMark Verified leaderboard docs70%

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

Data credits: MCPMark (eval-sys), mcpmark.ai Verified leaderboard. Single-run Pass@1 fractions shown as percentages.