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 | 71.9% | ok | 533/741 | $3.08 | 61 | 0.17h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 66.8% | ok | 495/741 | $3.97 | 66 | 0.18h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 3.6% | warn 1 | 27/742 | $3.00 | 59 | 0.16h | 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
71.9% from 533/741 ProgramBench-scored tests. Raw public eval statuses: failure: 290, passed: 634, skipped: 5.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch, terminal rendering mismatch.
d62734ec50dctests.test_cli_errors.test_cache_unknown_flag (failure)d62734ec50dctests.test_cache.test_cache_stats_size_units_kb (failure)d62734ec50dctests.test_export.test_export_mixed_case_flags (failure)d62734ec50dctests.test_global_args.test_unknown_subcommand (failure)d62734ec50dctests.test_history_stats.test_stats_help_after_invalid_flag (failure)d62734ec50dctests.test_cache_repo_integration.test_repo_clear_does_not_affect_cache_files (failure)
GPT 5.5 (xhigh) · 20260519T215046Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260518T232006Z
3.6% from 27/742 ProgramBench-scored tests. Raw public eval statuses: error: 1, failure: 797, not_run: 18, passed: 107, skipped: 5.
Agent trace: 136 shell command(s), 103 executable-observation command(s), 6 build command(s), 2 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch.
d62734ec50dctests.test_cache_repo_integration.test_repo_clear_then_cache_clear_both_empty (failure)d62734ec50dctests.test_cli_errors.test_history_help (failure)d62734ec50dctests.test_export.test_export_long_path (failure)d62734ec50dctests.test_cache_repo_integration.test_cache_stats_with_repos_present (failure)d62734ec50dctests.test_cli_errors.test_cache_list_empty (failure)d62734ec50dctests.test_export.test_export_help_flag (failure)
manifest · eval json · eval summary · agent summary · usage audit
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 | Anthropic | 91.3% | $1.74 | 71 |
| 2 | GPT 5.5 (xhigh) | OpenAI | 76.7% | $11.11 | 61 |
| 3 | GPT 5.5 (high) | OpenAI | 63.2% | $3.70 | 51 |
| 4 | GPT 5.4 | OpenAI | 63.1% | $0.25 | 10 |
| 5 | GPT 5.5 | OpenAI | 61.1% | $0.74 | 16 |
| 6 | Claude Haiku 4.5 | Anthropic | 56.3% | $0.34 | 93 |
| 7 | Claude Opus 4.7 (xhigh) | Anthropic | 50.3% | $4.73 | 133 |
| 8 | Gemini 3.1 Pro | 45.6% | $1.16 | 65 | |
| 9 | Gemini 3 Flash | 41.7% | $0.39 | 129 | |
| 10 | GPT 5 mini | OpenAI | 33.0% | $0.02 | 17 |
| 11 | Claude Sonnet 4.6 | Anthropic | 1.0% | $18.17 | 396 |
| 12 | GPT 5.4 mini | OpenAI | 1.0% | $0.05 | 24 |
| 13 | Claude Opus 4.6 | Anthropic | 0.0% | $4.14 | 242 |