344generated behavioral tests
97.4%official best score
91.6%Codex best score
3Codex result rows
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 | 91.6% | ok | 315/344 | $3.73 | 63 | 0.18h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 35.2% | ok | 121/344 | n/a | 94 | 0.17h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 34.9% | ok | 120/344 | n/a | 54 | 0.13h | 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
91.6% from 315/344 ProgramBench-scored tests. Raw public eval statuses: failure: 32, passed: 365, skipped: 3.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, terminal rendering mismatch.
f84528aebeb6eval.tests.test_edge_cases.test_config_with_empty_repo_path (failure)f84528aebeb6eval.tests.test_flag_combinations.test_unknown_flag_with_valid_flag (failure)f84528aebeb6eval.tests.test_config_creation.test_first_time_setup_with_no_config (failure)f84528aebeb6eval.tests.test_config_creation.test_config_dir_creation_pathway (failure)b31afd05cac9eval.tests.test_config_format.test_config_with_empty_string_path (failure)b31afd05cac9eval.tests.test_config_format.test_malformed_json_invalid_syntax (failure)
GPT 5.5 (xhigh) · 20260519T215046Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260518T232006Z
Public eval evidence has not been exported for this row yet.
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 (xhigh) | Anthropic | 97.4% | $19.76 | 260 |
| 2 | GPT 5.5 (high) | OpenAI | 93.6% | $2.56 | 36 |
| 3 | GPT 5.5 (xhigh) | OpenAI | 92.7% | $5.08 | 71 |
| 4 | GPT 5.5 | OpenAI | 84.0% | $1.10 | 20 |
| 5 | Claude Opus 4.6 | Anthropic | 78.8% | $5.67 | 195 |
| 6 | Claude Opus 4.7 | Anthropic | 66.6% | $2.50 | 104 |
| 7 | Claude Sonnet 4.6 | Anthropic | 42.7% | $10.77 | 364 |
| 8 | Gemini 3.1 Pro | 40.7% | $1.77 | 100 | |
| 9 | GPT 5.4 | OpenAI | 34.9% | $0.41 | 15 |
| 10 | Gemini 3 Flash | 30.5% | $0.21 | 66 | |
| 11 | GPT 5.4 mini | OpenAI | 26.5% | $0.05 | 11 |
| 12 | Claude Haiku 4.5 | Anthropic | 22.1% | $0.58 | 118 |
| 13 | GPT 5 mini | OpenAI | 16.3% | $0.04 | 16 |