Codex Results by Model
| # | Model | Run | Agent | Mode | Score | Eval | Tests | Est. cost | Calls | Wall | Evidence |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT 5.5 (xhigh) | 20260517T094700Z |
Codex /goal | Mini-SWE-compatible no internet | 84.5% | ok | 610/722 | $6.89 | 125 | 0.39h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 77.8% | ok | 562/722 | n/a | 61 | 0.17h | manifest · eval json · eval summary · agent summary · usage audit |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 73.3% | ok | 529/722 | n/a | 73 | 0.21h | 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
84.5% from 610/722 ProgramBench-scored tests. Raw public eval statuses: failure: 135, passed: 767, skipped: 4.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch.
ef68f5cfbae5eval.tests.test_autocomplete.test_autocomplete_mimes (failure)ef68f5cfbae5eval.tests.test_comprehensive_coverage.test_autocomplete_desktop_files_flag (failure)ef68f5cfbae5eval.tests.test_autocomplete.test_autocomplete_desktop_files (failure)ef68f5cfbae5eval.tests.test_final_coverage_push.test_launch_with_no_terminal_available (failure)ef68f5cfbae5eval.tests.test_more_edge_cases.test_autocomplete_empty_output (failure)ef68f5cfbae5eval.tests.test_set_get_unset.test_set_by_extension (failure)
GPT 5.5 (xhigh) · 20260518T232006Z
77.8% from 562/722 ProgramBench-scored tests. Raw public eval statuses: failure: 190, passed: 712, skipped: 4.
Agent trace: 111 shell command(s), 86 executable-observation command(s), 5 build command(s), 4 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch, exact version/output mismatch.
ef68f5cfbae5eval.tests.test_comprehensive_coverage.test_autocomplete_desktop_files_flag (failure)ef68f5cfbae5eval.tests.test_autocomplete.test_autocomplete_mimes (failure)ef68f5cfbae5eval.tests.test_autocomplete.test_autocomplete_desktop_files (failure)ef68f5cfbae5eval.tests.test_final_coverage_push.test_launch_with_no_terminal_available (failure)ef68f5cfbae5eval.tests.test_more_edge_cases.test_autocomplete_empty_output (failure)ef68f5cfbae5eval.tests.test_set_get_unset.test_get_json_output (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
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | GPT 5.5 (xhigh) | OpenAI | 91.4% | $7.47 | 68 |
| 2 | Claude Opus 4.7 | Anthropic | 90.7% | $7.87 | 169 |
| 3 | GPT 5.5 (high) | OpenAI | 89.3% | $3.01 | 45 |
| 4 | Claude Opus 4.7 (xhigh) | Anthropic | 85.5% | $14.90 | 287 |
| 5 | GPT 5.5 | OpenAI | 82.4% | $1.20 | 22 |
| 6 | Claude Sonnet 4.6 | Anthropic | 81.0% | $19.64 | 522 |
| 7 | Claude Opus 4.6 | Anthropic | 80.9% | $13.21 | 319 |
| 8 | Gemini 3.1 Pro | 78.0% | $1.41 | 121 | |
| 9 | GPT 5.4 | OpenAI | 65.4% | $0.25 | 10 |
| 10 | GPT 5.4 mini | OpenAI | 46.7% | $0.05 | 14 |
| 11 | Claude Haiku 4.5 | Anthropic | 31.9% | $0.74 | 156 |
| 12 | Gemini 3 Flash | 21.2% | $0.29 | 128 | |
| 13 | GPT 5 mini | OpenAI | 15.9% | $0.02 | 24 |