ProgramBench task detail

stacked-git__stgit.430027d

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.

1488generated behavioral tests
48.7%official best score
53.2%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 53.2% warn 1 821/1542 $3.88 58 0.20h manifest · eval json · eval summary · usage audit
2 GPT 5.5 (xhigh) 20260518T232006Z Codex /goal Paper prompt + /goal no internet 52.2% warn 1 802/1537 $3.84 58 0.20h manifest · eval json · eval summary · agent summary · usage audit
3 GPT 5.5 (xhigh) 20260519T215046Z Codex /goal Paper prompt + Goal contract no internet 51.3% ok 763/1488 $4.65 72 0.23h not exported

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

53.2% from 821/1542 ProgramBench-scored tests. Raw public eval statuses: failure: 873, not_run: 6, passed: 1444, skipped: 20.

Agent trace summary unavailable.

Likely miss class: behavioral mismatch.

  • debe624ffde1 tests.test_administration.test_completion_invalid_shell (failure)
  • debe624ffde1 tests.test_command_variants.test_completion_fish_all_commands (failure)
  • debe624ffde1 tests.test_additional_commands.test_new_patch_with_signoff (failure)
  • debe624ffde1 tests.test_administration.test_completion_bash (failure)
  • debe624ffde1 tests.test_branch_operations.test_branch_describe (failure)
  • debe624ffde1 tests.test_comprehensive_commands.test_new_with_committer_date_is_author_date (failure)

manifest · eval json · eval summary · usage audit

GPT 5.5 (xhigh) · 20260518T232006Z

52.2% from 802/1537 ProgramBench-scored tests. Raw public eval statuses: error: 1, failure: 850, not_run: 1, passed: 1466, skipped: 20.

Agent trace: 55 shell command(s), 33 executable-observation command(s), 6 build command(s), 2 blocked attempt(s). Raw Codex logs are not published.

Likely miss class: behavioral mismatch.

  • debe624ffde1 tests.test_command_variants.test_completion_fish_all_commands (failure)
  • debe624ffde1 tests.test_administration.test_completion_help (failure)
  • debe624ffde1 tests.test_administration.test_completion_bash (failure)
  • debe624ffde1 tests.test_command_variants.test_completion_bash_all_commands (failure)
  • debe624ffde1 tests.test_command_variants.test_squash_with_file_mode_changes (failure)
  • debe624ffde1 tests.test_additional_commands.test_show_with_patch_range (failure)

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

GPT 5.5 (xhigh) · 20260519T215046Z

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

Official ProgramBench Results by Model

#ModelProviderScoreCostCalls
1 GPT 5.5 (high) OpenAI 48.7% $4.62 44
2 GPT 5.5 (xhigh) OpenAI 46.4% $7.28 80
3 GPT 5.5 OpenAI 37.7% $1.13 18
4 Claude Opus 4.7 (xhigh) Anthropic 25.1% $10.31 162
5 GPT 5 mini OpenAI 20.0% $0.02 14
6 GPT 5.4 OpenAI 16.3% $0.23 8
7 Gemini 3 Flash Google 12.6% $0.24 114
8 Claude Opus 4.7 Anthropic 10.1% $0.56 25
9 Gemini 3.1 Pro Google 8.6% $1.18 94
10 Claude Haiku 4.5 Anthropic 8.3% $0.60 104
11 GPT 5.4 mini OpenAI 7.1% $0.02 9
12 Claude Sonnet 4.6 Anthropic 4.0% $31.22 581
13 Claude Opus 4.6 Anthropic 0.0% $1.93 213