Measured harness ledgerPublic result
DeepSeek V4 FlashOff-road Mud Game — DeepSeek V4 Flash 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
- $0.25
- Wall-clock
- 4h 35m 54.0s wall-clock
- Processed tokens
- 125.79M processed
- Record state
- partial_token_timing_artifact_and_build_ledger
Public summary
DeepSeek V4 Flash Max partial_token_timing_artifact_and_build_ledger ledger: 4h 35m 54.0s wall-clock, 125.79M processed, and $0.25 API-equivalent usage accounting, not an itemized subscription cash charge.
Run identity and stack
- Result ID: off-road-driving-game-threejs-deepseek-v4-flash-max
- Technical model: DeepSeek V4 Flash
- Provider: OpenCode Go
- Stack: OpenCode Go
- Stack: Technical model/configuration: DeepSeek V4 Flash
- Stack: Three.js vehicle workflow
- Stack: Harness v1 off-road prompt
Cost basis
- Visible output and reasoning are both output-priced. The fresh 2026-08-11 OpenCode Go rate entry is used; it supersedes an older higher rate snapshot. OpenCode Go subscription pricing is not divided into a marginal cash cost for this run.
Primary artifact integrity
- Kind: interactive-threejs-off-road-driving-game
- Path: artifacts/off-road-driving-game-threejs-deepseek-v4-flash-max/source/src/main.js
- SHA-256: 40bc4715c51f5012933ab6451367d10fc36bb2d7bee078e43c88d741c15016cb
Recorded caveats
- Wall-clock is end-to-end workflow latency including tool execution, installs, tests, browser checks, and idle gaps; it is not model-only compute.
- The run used nine benchmark-window user turns, so it is not a one-shot result.
- OpenCode splits one agentic turn into multiple assistant/API messages.
- Cache-read tokens are deeply discounted, so total processed tokens overstate cost.
- The public record excludes private OpenCode session/export data; source hashes and the stable benchmark-window hash are retained.
- The archive operator's source-level build and physics checks do not establish browser FPS or blind visual quality.
- No final RTX Pro 6000 render or blind-evaluation record was supplied.
Visible evidence gaps
- one-shot-compliant run
- browser and local-hardware identity
- local FPS
- final capture metadata
- 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.
