BuilderBuilder
GPT-6.1 Sol · Apple stem
The episode says all four pass, Opus on max finishes in 16 minutes 39 seconds, faster than GPT-6.1 Sol, and Astra is still fastest. Opus's display is 16m 39s. Astra max is 11m 05s. Sol is 25m 16s. Sonnet is 45m 59s. All four judge-pass. All four cycles-hero files match clips.public.json. HERO rendered capture Unaided model Open this round
Measured harness ledgerPublic result
GPT-6.1 Sol

Apple stem — GPT-6.1 Sol

Pinch an apple by the stem and lift it with a Wuji hand on an FR3 arm in SuperDex, without contacting the apple body.

Max reasoningHeadline result
Workflow cost
$0.82
Wall-clock
25m 16s
Processed tokens
3,382,863 processed (fresh input 101,335 + cache read 3,252,736 + cache creation 0 + output 28,792). Reasoning output 15,741 is a subset of output and is not added.
Record state
additive_harness_projection
Public summary

GPT-6.1 Sol Apple stem: 25m 16s, $0.82 API-equivalent estimate, 3,382,863 processed (fresh input 101,335 + cache read 3,252,736 + cache creation 0 + output 28,792). Reasoning output 15,741 is a subset of output and is not added..

Run identity and stack
  • Result ID: dex-apple-stem-superdex-gpt-6.1-sol-max-codex-2026-10-02
  • Technical model: gpt-6.1-sol
  • Provider: OpenAI Codex
  • Client: Codex CLI 0.160.0
  • Stack: SuperDex fingertip fixture
  • Stack: Harness apple-stem prompt
Cost basis
  • Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
  • Source: supplied_measurements.token_usage.total_processed at harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e.
Validation evidence
  • Result: PASS
  • Verification: Task verdict is judge_pass_independently_replayed. The replay gate is that status. Additive projection from harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e. Broad harness sync was not run. Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0. Source: supplied_measurements.token_usage.total_processed at harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e.
Recorded caveats
  • Judge pass. Cycles-hero sha256 matches. Opus at 16m 39s is faster than this Sol run, as the episode says.
  • Wall clock is supplied_measurements.wall_clock_display. Cost is supplied_measurements.cost.total API-equivalent. Charged amount is $0.
  • Source: supplied_measurements.token_usage.total_processed at harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e.
Visible evidence gaps
  • additive projection; broad harness sync was not run
  • source harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e
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 BuilderReturn to GPT-6.1 Sol
RemakeBenchResearch console