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 | 43.8% | ok | 216/493 | $5.76 | 91 | 0.29h | manifest · eval json · eval summary · agent summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260517T094700Z |
Codex /goal | Mini-SWE-compatible no internet | 42.0% | ok | 207/493 | $4.72 | 64 | 0.28h | manifest · eval json · eval summary · usage audit |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 38.9% | ok | 192/493 | $3.83 | 67 | 0.19h | 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) · 20260518T232006Z
43.8% from 216/493 ProgramBench-scored tests. Raw public eval statuses: failure: 417, passed: 470, skipped: 15.
Agent trace: 211 shell command(s), 96 executable-observation command(s), 95 build command(s), 2 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch.
bfd3ff740f11eval.tests.test_additional_coverage.test_check_deferInLoop_checker (failure)bfd3ff740f11eval.tests.test_additional_coverage.test_check_builtinShadowDecl_checker (failure)bfd3ff740f11eval.tests.test_check_parameters.test_configure_int_parameter (failure)bfd3ff740f11eval.tests.test_check_parameters.test_rangeValCopy_threshold (failure)bfd3ff740f11eval.tests.test_checker_detections.test_badCond_checker (failure)bfd3ff740f11eval.tests.test_checker_detections.test_exitAfterDefer_checker (failure)
manifest · eval json · eval summary · agent summary · usage audit
GPT 5.5 (xhigh) · 20260517T094700Z
42.0% from 207/493 ProgramBench-scored tests. Raw public eval statuses: failure: 433, passed: 454, skipped: 15.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch.
bfd3ff740f11eval.tests.test_check_basic.test_check_file_with_issue (failure)bfd3ff740f11eval.tests.test_additional_coverage.test_check_deferInLoop_checker (failure)bfd3ff740f11eval.tests.test_additional_coverage.test_check_builtinShadowDecl_checker (failure)bfd3ff740f11eval.tests.test_check_parameters.test_configure_int_parameter (failure)bfd3ff740f11eval.tests.test_checker_detections.test_badCond_checker (failure)bfd3ff740f11eval.tests.test_checker_detections.test_sloppyTypeAssert_checker (failure)
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 | 56.8% | $9.11 | 82 |
| 2 | GPT 5.5 (high) | OpenAI | 49.7% | $3.64 | 43 |
| 3 | Claude Sonnet 4.6 | Anthropic | 41.6% | $34.12 | 677 |
| 4 | Claude Opus 4.6 | Anthropic | 35.1% | $20.28 | 360 |
| 5 | GPT 5.5 | OpenAI | 34.1% | $1.19 | 21 |
| 6 | GPT 5.4 | OpenAI | 30.4% | $0.49 | 14 |
| 7 | Claude Opus 4.7 | Anthropic | 28.2% | $4.00 | 134 |
| 8 | Claude Opus 4.7 (xhigh) | Anthropic | 27.8% | $4.79 | 122 |
| 9 | Gemini 3 Flash | 26.0% | $0.21 | 58 | |
| 10 | Gemini 3.1 Pro | 23.9% | $1.15 | 111 | |
| 11 | Claude Haiku 4.5 | Anthropic | 21.7% | $0.46 | 89 |
| 12 | GPT 5.4 mini | OpenAI | 3.9% | $0.04 | 7 |
| 13 | GPT 5 mini | OpenAI | 0.6% | $0.02 | 20 |