Off-road Mud Game — GPT-6 Astra Max

Measured harness ledgerPublic result

Off-road Mud Game — GPT-6 Astra Max

Build a one-request procedural off-road driving game with vehicle physics, independent suspension, streamed terrain, mud feedback, and multiple cameras.

Max reasoningHeadline result

Workflow cost: $274.71
Wall-clock: 13h 57m 41.1s wall-clock
Processed tokens: 195.96M processed
Record state: partial_cumulative_root_snapshot_with_clean_build_and_model_supplied_runtime_evidence

Public summary

GPT-6 Astra Max partial_cumulative_root_snapshot_with_clean_build_and_model_supplied_runtime_evidence ledger: 13h 57m 41.1s wall-clock, 195.96M processed, and $274.71 Standard API-equivalent estimate from the supplied root Codex receipt, not an actual subscription charge or all-agent spend..

Run identity and stack

Result ID: off-road-driving-game-threejs-gpt-6-astra-max
Technical model: gpt-6-astra
Provider: OpenAI Codex
Stack: OpenAI Codex
Stack: Technical model/configuration: gpt-6-astra
Stack: Three.js vehicle workflow
Stack: Harness v1 off-road prompt

Primary artifact integrity

Kind: interactive-threejs-off-road-driving-game
Path: artifacts/off-road-driving-game-threejs-gpt-6-astra-max/source/src/main.js
SHA-256: 9debf02c80a8f4534a9837d1ac9514f515ef3beeb7768509cff4dc4a4b9b8fbf

Recorded caveats

The cost, tokens, and wall-clock are a cumulative root-transcript snapshot through the metrics request, not a strict completed-generation-only measurement.
Subagent logs are excluded to avoid inherited-context double counting, so the displayed figures are not a complete all-agent spend total.
Wall-clock includes tools and idle gaps; output includes hidden reasoning, visible prose/code, and tool-call JSON.
Cache reads are discounted, so processed-token volume overstates effective cost.
The archive operator reproduced a clean build but did not independently drive the game, measure FPS, or judge visual or 3D-asset quality. Submitted runtime and visual evidence is labelled model-supplied.
The submitted driving receipt reports 31.95 FPS on an Apple M1 Pro Balanced setting, not 60 FPS or a standardized desktop-GPU result.
The submitted report documents an explicit eight-round limit and remaining visual limitations.
No blind evaluation is recorded.

Visible evidence gaps

strict completed-generation-only all-agent token and timing ledger
independent browser gameplay replay through real input
standardized fixed-hardware FPS measurement
independent visual and 3D asset-quality acceptance
blind-evaluation record

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.

Browse Builder Return to GPT-6 Astra