BuilderBuilder
Claude Sonnet 5.5 · Two-handed puzzle
The episode says the Claude models fail and both GPT models pass. The unaided max ledgers agree: Astra and GPT-6.1 Sol judge-pass, Opus and Sonnet judge-fail. Opus's failure includes the judge result, not only the narration about the simulation breaking. HERO rendered capture Unaided model Open this round
Measured harness ledgerPublic result
Claude Sonnet 5.5

Two-handed puzzle — Claude Sonnet 5.5

Find a collision-free rigid motion that unlocks two interlocked wire claws using pure geometry.

Max reasoningHeadline result
Workflow cost
$13.09
Wall-clock
1h 39m 54s
Processed tokens
41,721,200 processed (fresh input 306 + cache read 40,976,571 + cache creation 425,075 + output including reasoning 319,248). Visible output 123,040 and reasoning output 196,208 are inside that output figure and are not added again.
Record state
additive_harness_projection
Public summary

Claude Sonnet 5.5 Two-handed puzzle: 1h 39m 54s, $13.09 API-equivalent estimate, 41,721,200 processed (fresh input 306 + cache read 40,976,571 + cache creation 425,075 + output including reasoning 319,248). Visible output 123,040 and reasoning output 196,208 are inside that output figure and are not added again..

Run identity and stack
  • Result ID: claw-unlock-geometry-sonnet-5.5-max-2026-09-30
  • Technical model: claude-sonnet-5-5
  • Provider: Anthropic Claude Code
  • Client: Claude CLI 2.1.286
  • Stack: python-fcl geometry fixture
  • Stack: Harness claw-unlock prompt
Cost basis
  • Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0.
  • Source: supplied_measurements.token_usage.total_processed at harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e.
Validation evidence
  • Result: FAIL
  • Verification: Task verdict is judge_fail_independently_replayed. Independent replay validation is PASS and does not replace the task verdict. Plotted judgment stays FAIL. Additive projection from harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e. Broad harness sync was not run. Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge is $0. Source: supplied_measurements.token_usage.total_processed at harness commit 9b7c614dbb5103cc373fdf6275b5c4efc10d8f0e.
Recorded caveats
  • Judge fail. Cycles-hero sha256 matches. The hero does not change the fail.
  • Wall clock is supplied_measurements.wall_clock_display. Cost is the recorded list-price total. Subscription charge 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 Claude Sonnet 5.5
RemakeBenchResearch console