ProgramBench task detail

hatoo__oha.8dc6349

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.

899generated behavioral tests
78.6%official best score
83.9%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 83.9% ok 754/899 $6.21 107 0.28h manifest · eval json · eval summary · usage audit
2 GPT 5.5 (xhigh) 20260518T232006Z Codex /goal Paper prompt + /goal no internet 69.0% ok 620/899 n/a 51 0.14h manifest · eval json · eval summary · agent summary · usage audit
3 GPT 5.5 (xhigh) 20260519T215046Z Codex /goal Paper prompt + Goal contract no internet 57.8% ok 520/899 n/a 53 0.19h 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

83.9% from 754/899 ProgramBench-scored tests. Raw public eval statuses: error: 5, failure: 185, passed: 899, skipped: 6.

Agent trace summary unavailable.

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

  • dd30c5f58032 tests.test_basic.test_missing_url_error (failure)
  • dd30c5f58032 tests.test_debug.test_db_url_option (failure)
  • dd30c5f58032 tests.test_edge_cases.test_request_count_as_float (failure)
  • dd30c5f58032 tests.test_errors.test_invalid_regex_url (failure)
  • dd30c5f58032 tests.test_errors.test_conflicting_body_options (failure)
  • dd30c5f58032 tests.test_errors.test_invalid_http_version (failure)

manifest · eval json · eval summary · usage audit

GPT 5.5 (xhigh) · 20260518T232006Z

69.0% from 620/899 ProgramBench-scored tests. Raw public eval statuses: error: 11, failure: 341, passed: 737, skipped: 6.

Agent trace: 70 shell command(s), 42 executable-observation command(s), 8 build command(s), 1 blocked attempt(s). Raw Codex logs are not published.

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

  • dd30c5f58032 tests.test_basic.test_invalid_url_error (failure)
  • dd30c5f58032 tests.test_debug.test_db_url_option (failure)
  • dd30c5f58032 tests.test_advanced.test_mixed_success_failure (failure)
  • dd30c5f58032 tests.test_edge_cases.test_request_count_as_float (failure)
  • dd30c5f58032 tests.test_basic.test_missing_url_error (failure)
  • dd30c5f58032 tests.test_errors.test_invalid_url_format (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 (xhigh) OpenAI 78.6% $5.02 41
2 Claude Sonnet 4.6 Anthropic 72.5% $19.74 435
3 GPT 5.5 (high) OpenAI 70.6% $2.55 28
4 GPT 5.5 OpenAI 68.3% $1.16 13
5 Claude Opus 4.7 (xhigh) Anthropic 61.8% $4.10 75
6 GPT 5.4 OpenAI 52.8% $0.28 8
7 Claude Opus 4.7 Anthropic 45.8% $1.40 39
8 Gemini 3.1 Pro Google 34.1% $0.96 43
9 Gemini 3 Flash Google 26.4% $0.29 55
10 GPT 5.4 mini OpenAI 15.8% $0.03 7
11 GPT 5 mini OpenAI 12.7% $0.02 8
12 Claude Opus 4.6 Anthropic 0.0% $2.89 187
13 Claude Haiku 4.5 Anthropic 0.0% $0.93 99