BuilderBuilder
Measured harness ledgerPublic result
Kimi K3

Fighter Game v1 — Kimi K3 Unreported

Improve the complete arena environment of a fixed Three.js fighting game through Blender MCP without changing the fighters, combat, HUD, effects, or audio.

Unreported reasoningHeadline result
Workflow cost
≥$2.91
Wall-clock
31m 15.6s wall-clock
Processed tokens
6.59M processed
Record state
artifact_independent_boot_ledger_final_turn_metrics_only
Public summary

Kimi K3 Unreported artifact_independent_boot_ledger_final_turn_metrics_only ledger: 31m 15.6s wall-clock, 6.59M processed, and ≥$2.91 Independently recomputed API-list-price equivalent for the measured token mix; not an itemized subscription charge (lower bound).

Run identity and stack
  • Result ID: courtyard-arena-blender-mcp-kimi-k3-max
  • Technical model: k3
  • Provider: openai-compatible
  • Client: Kimi Code 0.34.0
  • Stack: openai-compatible
  • Stack: Kimi Code 0.34.0
  • Stack: Technical model/configuration: k3
  • Stack: Blender MCP
  • Stack: Three.js fighting-game fixture
  • Stack: Harness v1 frozen prompt
Cost basis
  • FINAL TURN ONLY — not the cost of the run. See measurement_window_caveat.
Primary artifact integrity
  • Kind: manifest-verified-artifact
  • Path: artifacts/courtyard-arena-blender-mcp-kimi-k3-max/source/web/models/arena.glb
  • SHA-256: 49ebeecdb092e418a3f665b825ffc4c455d02990be21c97f67bd905f5ca2363e
Validation evidence
  • Result: PASS
Recorded caveats
  • THE METRICS COVER ONE TURN, NOT THE RUN. The receipt validates cleanly but its window spans only the final turn; 167 delivered files predate the window against 22 inside it, and the session's own permission events are stamped about three hours earlier. Tokens, wall-clock, and cost are lower bounds on a closing turn.
  • Wall-clock includes tools, installs, browser checks, approval waits, and idle time — not model-only compute.
  • One user turn can contain many billable model requests.
  • Kimi Code does not emit a separate reasoning-token count in the observed format, so reasoning is folded into the 10,074 output tokens and cannot be separated.
  • Cache reads are discounted, so total processed tokens overstate cost.
  • The Allegretto plan does not itemize a per-run cash charge; the $2.91 is an API-list-price equivalent for the measured turn only.
  • No quota evidence exists — there is no /usage capture in the session.
  • This is an environment-improvement task over a fixed starter, so its figures are not comparable to the archive's from-scratch build rows.
  • The archive is a delta over the committed fixture, not a standalone runnable tree.
  • Roughly 124 MB of baseline and iteration captures were produced; only the final and clay sets are archived.
  • No frame-rate receipt exists — neither the model nor the operator recorded a valid FPS sample.
  • No blind evaluation and no scored side-by-side against the reference run or the Opus 5 candidate are recorded.
  • The private session ZIP named in receipt.sha256 is withheld from the public archive by its own privacy label.
Visible evidence gaps
  • a whole-session metrics receipt covering every turn of the run
  • blind-evaluation record
  • standardized FPS and hitch receipt under sustained play
  • scored side-by-side against the reference run's judging anchors
Builder test available

This result is part of a Builder test. Open it for the exact prompt and any released projects, RemakeBench Harness workflows and production skills. Public proof and known evidence gaps stay visible here.

  • Fighter Game v1 · Builder projects v1
View Fighter Game v1 · Builder projects v1Return to Kimi K3
RemakeBenchResearch console