Codex Results by Model
| # | Model | Run | Agent | Mode | Score | Eval | Tests | Est. cost | Calls | Wall | Evidence |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 52.2% | ok | 611/1171 | n/a | 54 | 0.21h | manifest · eval json · eval summary · agent summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 52.1% | ok | 610/1171 | n/a | 79 | 0.27h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260517T094700Z |
Codex /goal | Mini-SWE-compatible no internet | 49.4% | ok | 579/1171 | $8.63 | 139 | 0.32h | manifest · eval json · eval 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) · 20260518T232006Z
52.2% from 611/1171 ProgramBench-scored tests. Raw public eval statuses: failure: 581, passed: 681, skipped: 4.
Agent trace: 95 shell command(s), 48 executable-observation command(s), 7 build command(s), 1 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch, exact version/output mismatch.
e68dc00dc7d5eval.tests.test_authentication.test_basic_auth (failure)e68dc00dc7d5eval.tests.test_file_operations.test_stdin_as_body (failure)e68dc00dc7d5eval.tests.test_authentication.test_basic_auth_encoded (failure)e68dc00dc7d5eval.tests.test_authentication.test_bearer_auth_with_long_token (failure)e68dc00dc7d5eval.tests.test_authentication.test_auth_short_form (failure)e68dc00dc7d5eval.tests.test_file_operations.test_header_value_from_file (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.
GPT 5.5 (xhigh) · 20260517T094700Z
49.4% from 579/1171 ProgramBench-scored tests. Raw public eval statuses: failure: 610, passed: 652, skipped: 4.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch.
e68dc00dc7d5eval.tests.test_file_operations.test_header_value_from_file (failure)e68dc00dc7d5eval.tests.test_generation.test_generate_zsh_completion (failure)e68dc00dc7d5eval.tests.test_generation.test_generate_powershell_completion (failure)e68dc00dc7d5eval.tests.test_generation.test_generate_nushell_completion (failure)e68dc00dc7d5eval.tests.test_generation.test_generate_bash_completion (failure)e68dc00dc7d5eval.tests.test_generation.test_generate_elvish_completion (failure)
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6 | Anthropic | 50.0% | $13.67 | 284 |
| 2 | GPT 5.5 (high) | OpenAI | 48.5% | $2.92 | 27 |
| 3 | GPT 5.5 (xhigh) | OpenAI | 43.5% | $10.20 | 89 |
| 4 | GPT 5.5 | OpenAI | 42.2% | $1.18 | 17 |
| 5 | Claude Opus 4.7 (xhigh) | Anthropic | 38.6% | $3.96 | 82 |
| 6 | GPT 5.4 | OpenAI | 36.6% | $0.25 | 7 |
| 7 | Claude Sonnet 4.6 | Anthropic | 33.6% | $45.79 | 702 |
| 8 | Gemini 3.1 Pro | 24.7% | $0.84 | 54 | |
| 9 | Claude Haiku 4.5 | Anthropic | 19.8% | $1.42 | 156 |
| 10 | Gemini 3 Flash | 16.7% | $0.23 | 73 | |
| 11 | Claude Opus 4.7 | Anthropic | 15.5% | $1.24 | 43 |
| 12 | GPT 5 mini | OpenAI | 9.4% | $0.02 | 13 |
| 13 | GPT 5.4 mini | OpenAI | 1.3% | $0.07 | 14 |