---
title: CyBench x Kimi K3 — Transcript Runs (Inspect View)
emoji: 🕵️
sdk: static
---

# CyBench × Kimi K3 — all runs (Inspect View bundle)

Inspect View bundle of every CyBench run of Moonshot AI's Kimi K3 (2.8T MXFP4 MoE,
routed via OpenRouter) collected for the APAC transcript-risk study. One `.eval`
per run batch:

- `pilot_A` / `pilot_B` — routing-config pilot (Moonshot-hosted reasoning-on vs
  Together reasoning-off)
- `fullA_seedN_unguided` / `fullA_seedN_subtask` — full benchmark passes
  (40 tasks × 3 seeds × both modes, paper config: 15/5 iterations, 6k/2k tokens)
- `retry_seedN_*` — retry passes for tasks whose environments needed the
  archived-distro repair (see dataset card)

Companion artifacts:
- Transcripts (native CyBench JSON): `ajay-citadel/cybench-kimi-k3-transcripts`
- Harness + patches provenance: see dataset card and PATCHES.md therein

Logs are added as passes complete; refresh to see new batches.