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 | 87.6% | ok | 621/709 | $3.66 | 59 | 0.23h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 87.4% | ok | 620/709 | n/a | 59 | 0.20h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 81.9% | ok | 581/709 | n/a | 50 | 0.15h | 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
87.6% from 621/709 ProgramBench-scored tests. Raw public eval statuses: failure: 102, passed: 891, skipped: 9.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch.
c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_launches_with_tmux (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_with_current_directory (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_displays_percentages (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_multiple_paths (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_i_alias (failure)c2599e7591b2eval.tests.test_stress_coverage.test_aggregate_with_all_format_options (failure)
GPT 5.5 (xhigh) · 20260519T215046Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260518T232006Z
81.9% from 581/709 ProgramBench-scored tests. Raw public eval statuses: failure: 155, passed: 838, skipped: 9.
Agent trace: 141 shell command(s), 118 executable-observation command(s), 4 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.
c2599e7591b2eval.tests.test_error_paths.test_invalid_arguments (failure)c2599e7591b2eval.tests.test_error_conditions.test_invalid_thread_count (failure)c2599e7591b2eval.tests.test_extreme_coverage.test_completions_detailed_output (failure)c2599e7591b2eval.tests.test_extreme_coverage.test_symlink_combinations (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_with_no_entry_check (failure)c2599e7591b2eval.tests.test_interactive_tmux.test_interactive_with_subdirectories (failure)
manifest · eval json · eval summary · agent summary · usage audit
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | GPT 5.5 (xhigh) | OpenAI | 93.1% | $11.00 | 74 |
| 2 | Claude Opus 4.7 (xhigh) | Anthropic | 90.8% | $10.03 | 161 |
| 3 | Claude Opus 4.6 | Anthropic | 86.9% | $10.56 | 277 |
| 4 | GPT 5.5 (high) | OpenAI | 84.6% | $2.81 | 29 |
| 5 | Claude Sonnet 4.6 | Anthropic | 83.9% | $22.21 | 502 |
| 6 | GPT 5.5 | OpenAI | 78.3% | $1.28 | 25 |
| 7 | Claude Opus 4.7 | Anthropic | 70.8% | $2.82 | 80 |
| 8 | Claude Haiku 4.5 | Anthropic | 49.1% | $1.27 | 137 |
| 9 | GPT 5.4 | OpenAI | 40.5% | $0.17 | 7 |
| 10 | GPT 5 mini | OpenAI | 39.1% | $0.04 | 22 |
| 11 | GPT 5.4 mini | OpenAI | 26.0% | $0.05 | 11 |
| 12 | Gemini 3.1 Pro | 4.2% | $1.87 | 131 | |
| 13 | Gemini 3 Flash | 0.0% | $0.31 | 100 |