ProgramBench task detail

dandavison__delta.acd758f

Task-level results for this Codex /goal scaffold. ProgramBench baseline context is cached from the official task page; Codex scored tests are after active-branch and ignored-test filtering.

950generated behavioral tests
38.9%official best score
35.9%Codex best score
3Codex result rows

Codex Results by Model

#ModelRunAgentModeScoreEvalTestsEst. costCallsWallEvidence
1 GPT 5.5 (xhigh) 20260517T094700Z Codex /goal Mini-SWE-compatible no internet 35.9% ok 341/950 $4.56 46 0.20h manifest · eval json · eval summary · usage audit
2 GPT 5.5 (xhigh) 20260519T215046Z Codex /goal Paper prompt + Goal contract no internet 16.2% ok 154/950 n/a 80 0.26h not exported
3 GPT 5.5 (xhigh) 20260518T232006Z Codex /goal Paper prompt + /goal no internet 0.7% ok 7/950 n/a 81 0.19h manifest · eval json · eval summary · agent summary · usage audit

Why Scores Differ

Official ProgramBench rows are the public mini-SWE-agent submissions. GoalBench rows are separate Codex /goal submissions. Same model label does not mean the same agent, prompt, tool loop, or generated implementation. A GoalBench row is resolved only when the ProgramBench-scored pass rate is exactly 100%.

GPT 5.5 (xhigh) · 20260517T094700Z

35.9% from 341/950 ProgramBench-scored tests. Raw public eval statuses: failure: 662, passed: 526.

Agent trace summary unavailable.

Likely miss class: behavioral mismatch.

  • 3c702ddd0c23 tests.test_ansi_iterator.test_rgb_foreground_color (failure)
  • 3c702ddd0c23 tests.test_ansi_gaps.test_csi_clear_sequences_stripped (failure)
  • 3c702ddd0c23 tests.test_ansi_iterator.test_mixed_text_and_ansi (failure)
  • 3c702ddd0c23 tests.test_blame.test_blame_format_no_placeholders (failure)
  • 3c702ddd0c23 tests.test_bat_output_gaps.test_paging_with_color_preservation (failure)
  • 3c702ddd0c23 tests.test_ansi_gaps.test_accumulated_attributes (failure)

manifest · eval json · eval summary · usage audit

GPT 5.5 (xhigh) · 20260519T215046Z

Public eval evidence has not been exported for this row yet.

GPT 5.5 (xhigh) · 20260518T232006Z

0.7% from 7/950 ProgramBench-scored tests. Raw public eval statuses: failure: 1151, passed: 37.

Agent trace: 151 shell command(s), 123 executable-observation command(s), 9 build command(s), 0 blocked attempt(s). Raw Codex logs are not published.

Likely miss class: behavioral mismatch, exact version/output mismatch.

  • 3c702ddd0c23 tests.test_ansi_gaps.test_accumulated_attributes (failure)
  • 3c702ddd0c23 tests.test_ansi_gaps.test_csi_clear_sequences_stripped (failure)
  • 3c702ddd0c23 tests.test_ansi_iterator.test_mixed_text_and_ansi (failure)
  • 3c702ddd0c23 tests.test_bat_output_gaps.test_delta_as_pager_fatal_error (failure)
  • 3c702ddd0c23 tests.test_ansi_iterator.test_rgb_foreground_color (failure)
  • 3c702ddd0c23 tests.test_diff_stat.test_narrow_stat_width (failure)

manifest · eval json · eval summary · agent summary · usage audit

Official ProgramBench Results by Model

#ModelProviderScoreCostCalls
1 GPT 5.5 (xhigh) OpenAI 38.9% $10.75 65
2 Claude Sonnet 4.6 Anthropic 37.3% $34.08 470
3 GPT 5.5 (high) OpenAI 31.8% $6.94 57
4 GPT 5.5 OpenAI 30.1% $0.96 14
5 Claude Opus 4.7 (xhigh) Anthropic 28.1% $1.19 48
6 Claude Opus 4.7 Anthropic 23.1% $0.35 22
7 Gemini 3 Flash Google 21.3% $0.20 52
8 GPT 5 mini OpenAI 20.7% $0.03 14
9 GPT 5.4 OpenAI 17.8% $0.23 7
10 Claude Haiku 4.5 Anthropic 10.1% $1.28 166
11 GPT 5.4 mini OpenAI 5.4% $0.05 8
12 Claude Opus 4.6 Anthropic 0.8% $8.32 236
13 Gemini 3.1 Pro Google 0.0% $1.70 120