Measured harness ledgerPublic result
Gemini 3.8 FlashEiffel Tower Drawing — Gemini 3.8 Flash High
Draw the Eiffel Tower with a friction-held pencil on a Marvin arm and Wuji hand in MuJoCo.
High reasoningHeadline result
- Workflow cost
- $2.72
- Wall-clock
- 8m 25s wall-clock
- Processed tokens
- 7.72M processed
- Record state
- judge_pass_independently_replayed
Public summary
Gemini 3.8 Flash High judge_pass_independently_replayed ledger: 8m 25s wall-clock, 7.72M processed, and $2.72 API-equivalent list price, not a marginal subscription cash charge.
Run identity and stack
- Result ID: dex-draw-mujoco-gemini-3.8-flash-high-antigravity-2026-09-16
- Technical model: Gemini 3.8 Flash
- Provider: Google Antigravity
- Client: Antigravity CLI 1.1.5 (headless print mode)
- Stack: Google Antigravity
- Stack: Antigravity CLI 1.1.5 (headless print mode)
- Stack: Technical model/configuration: Gemini 3.8 Flash
- Stack: MuJoCo pencil-drawing fixture
- Stack: Harness drawing prompt
- Stack: Requested tool profile: python-mujoco-headless
Cost basis
- actual marginal subscription charge usd: $0.00 USD.
- Introductory 2026 API-equivalent alternative: $1.36.
Primary artifact integrity
- Kind: open-loop-1khz-actuator-log
- Path: artifacts/dex-draw-mujoco-gemini-3.8-flash-high-antigravity-2026-09-16/evidence/model-supplied/ctrl.npy
- SHA-256: c16e78101f93ae5f30e234709937bc0685882679bf7051fa1848d0b707fe5f28
Validation evidence
- Result: pass
- Path: artifacts/dex-draw-mujoco-gemini-3.8-flash-high-antigravity-2026-09-16/validation/judge-replay.public.json
- SHA-256: 165df64520aecd8cf5add163260500f119cac2161819b459ce9e3a2bdf29a4df
- Validator SHA-256: 5c7af1a85044163afefca6abcd1ea9a19b109bf045ed8be40b4d5796098b738d
Recorded caveats
- The judge measures task completion only (contact-gated ink against the given polyline, runtime limits); there is no visual quality or blind evaluation for this task.
- The replay and progression clips are rendered by the operator's viewer from the judge's trajectory export and ink stream with a fixed camera; they are presentation evidence, not a measurement.
- usedThreeFingers / threeFingerFraction are reported, never gating.
- Wall clock is end-to-end workflow latency including the solver's own simulation and judge runs, not model-only compute, and all four runs of this batch shared one Mac.
- The given harness exposes the judge's own simulate/rasterise/score pipeline, so the solver could self-score in-process; wall clock and cost are NOT comparable with the dex-cube-turn-mujoco cohorts.
- This run is on fixture revision v1.1 (harness simulation archiving only). Verdicts stay comparable with the three v1 runs; scene_sha256 differs because it hashes harness.py.
- Cost is API-equivalent on two published bases (introductory 2026 and standard); the subscription's marginal charge was $0 and print mode exposes no quota delta.
- This run used the newer gemini-3.8-flash-high model in headless print mode rather than the operator's frozen Gemini 3.6 Flash TUI stack; disclosed as a stack deviation.
- The conversation id is withheld per precedent; raw usage receipts, the stream-json transcript, the CLI log and the conversation database are committed by hash only, and provenance/run-notes.md is a sanitized copy.
- One of the archived judge attempts is a paper-tearing failure (tip force averaged 3.021 N over 20 ms); it is the opening segment of the progression clip.
Visible evidence gaps
- hardware and interpreter receipt from the solver session (only the operator's replay environment is recorded)
Public result only
This result keeps its public summary and evidence, but it does not currently have a matching Builder test with prompts, projects, Harness workflows or skills.