Measured harness ledgerPublic result
GPT-6 Astra (low) planner + Jev 1.13 actor

Dex Cube Turn — GPT-6 Astra (low) planner + Jev 1.13 actor Unreported

Turn one face of a Rubik's cube a quarter turn by finger contact with two dexterous hands in MuJoCo.

Unreported reasoningHeadline result
Workflow cost
$0.91
Wall-clock
Not recorded
Processed tokens
Not recorded
Record state
judge_pass_independently_replayed
Public summary

GPT-6 Astra (low) planner + Jev 1.13 actor Unreported judge_pass_independently_replayed ledger: Not recorded wall-clock, Not recorded, and $0.91 Recorded provider cost.

Run identity and stack
  • Result ID: dex-cube-turn-mujoco-astra-planner-jev-2026-09-21
  • Technical model: GPT-6 Astra (low) planner + Jev 1.13 actor
  • Provider: OpenAI Codex (planner) + TypeSafe via OpenRouter (Jev)
  • Stack: OpenAI Codex (planner) + TypeSafe via OpenRouter (Jev)
  • Stack: Technical model/configuration: GPT-6 Astra (low) planner + Jev 1.13 actor
  • Stack: MuJoCo dexterous-hand fixture
  • Stack: Harness cube-turn prompt
Validation evidence
  • Path: artifacts/dex-cube-turn-mujoco-astra-planner-jev-2026-09-21/validation/judge-replay.public.json
  • SHA-256: 95666054d7799113eee34509b14e53554e0aed695a3027993d5e4599d9f08312
Recorded caveats
  • NOT COMPARABLE WITH THE CODING-AGENT COHORTS OR WITH BARE DECISION-LOOP ROWS. This is a jev-astra-hybrid run: GPT-6 Astra and Jev 1.13 were coupled under a protocol that is not the one-shot locked-prompt coding-agent track and not the two-call-per-cycle decision-loop track. cohort_eligible is false. Harry's criterion (same verdict as Astra-alone AND cheaper on API-equivalent) is recorded in the track README; it is not a league-table standing.
  • Astra-alone (low) coding-agent baseline is registered (2026-09-21). Arm C is cheaper than both Astra-max and Astra-low on this task.
  • A Blender Cycles HERO RE-RENDER of this judged replay is pinned in provenance/clips.public.json as presentation: cycles-hero. Poses identical, visual layer only; pipeline ~/hero-render; one RTX 4090 session, 130 min, $1.61 + $0.04 discarded pod. It is not a new run, adds no measurement and cannot change a verdict. It is not committed; it is pinned by name, sha256, bytes, frames and duration.
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.

RemakeBenchResearch console