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 | 91.5% | ok | 551/602 | $3.63 | 49 | 0.21h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 88.0% | ok | 530/602 | $2.93 | 51 | 0.18h | manifest · eval json · eval summary · agent summary · usage audit |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 6.5% | ok | 39/602 | $4.15 | 62 | 0.26h | 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
91.5% from 551/602 ProgramBench-scored tests. Raw public eval statuses: failure: 51, passed: 657, skipped: 1.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch.
d161b26b57a6eval.tests.test_input_handling.test_nonexistent_file_error (failure)d161b26b57a6eval.tests.test_input_handling.test_invalid_nix_syntax_error (failure)d161b26b57a6eval.tests.test_output_formats.test_json_output_parse_error (failure)d161b26b57a6eval.tests.test_special_features.test_case_insensitive_nix_extension (failure)d161b26b57a6eval.tests.test_special_features.test_lambda_arg_underscore_convention (failure)8a961ab581f3eval.tests.test_deadnix.test_combination_of_multiple_flags (failure)
GPT 5.5 (xhigh) · 20260518T232006Z
88.0% from 530/602 ProgramBench-scored tests. Raw public eval statuses: failure: 72, passed: 636, skipped: 1.
Agent trace: 50 shell command(s), 30 executable-observation command(s), 3 build command(s), 1 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch.
d161b26b57a6eval.tests.test_output_formats.test_no_color_environment_variable (failure)d161b26b57a6eval.tests.test_input_handling.test_invalid_nix_syntax_error (failure)d161b26b57a6eval.tests.test_special_features.test_lambda_arg_underscore_convention (failure)d161b26b57a6eval.tests.test_output_formats.test_json_output_file_not_found (failure)d161b26b57a6eval.tests.test_special_features.test_case_insensitive_nix_extension (failure)d161b26b57a6eval.tests.test_output_formats.test_json_output_parse_error (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 | 91.7% | $9.41 | 58 |
| 2 | Claude Opus 4.7 (xhigh) | Anthropic | 90.0% | $19.53 | 167 |
| 3 | Claude Opus 4.7 | Anthropic | 85.5% | $9.59 | 142 |
| 4 | Claude Sonnet 4.6 | Anthropic | 81.4% | $12.78 | 294 |
| 5 | GPT 5.5 | OpenAI | 80.1% | $1.74 | 20 |
| 6 | Gemini 3.1 Pro | 70.1% | $1.71 | 81 | |
| 7 | Gemini 3 Flash | 57.5% | $0.27 | 57 | |
| 8 | Claude Haiku 4.5 | Anthropic | 44.9% | $1.23 | 122 |
| 9 | GPT 5.4 | OpenAI | 44.5% | $0.40 | 9 |
| 10 | GPT 5 mini | OpenAI | 23.1% | $0.05 | 8 |
| 11 | GPT 5.5 (high) | OpenAI | 6.5% | $4.79 | 43 |
| 12 | GPT 5.4 mini | OpenAI | 2.0% | $0.05 | 29 |
| 13 | Claude Opus 4.6 | Anthropic | 0.0% | $3.29 | 258 |