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 | 61.1% | ok | 469/767 | n/a | 81 | 0.17h | manifest · eval json · eval summary · agent summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260517T094700Z |
Codex /goal | Mini-SWE-compatible no internet | 40.4% | ok | 310/767 | $3.83 | 63 | 0.24h | manifest · eval json · eval summary · usage audit |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 37.0% | ok | 284/767 | n/a | 52 | 0.17h | 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
61.1% from 469/767 ProgramBench-scored tests. Raw public eval statuses: failure: 314, passed: 661, skipped: 3.
Agent trace: 116 shell command(s), 77 executable-observation command(s), 4 build command(s), 2 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch.
e9e8360a457ceval.tests.test_oranda_comprehensive.test_no_arguments_shows_help (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_generate_ci_help (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_serve_without_build (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_generate_ci_requires_tty (failure)797ec2fcc5dbeval.tests.test_basic_invocation.test_no_args_shows_help (failure)797ec2fcc5dbeval.tests.test_cargo_detection.test_build_cargo_and_config_preference (failure)
manifest · eval json · eval summary · agent summary · usage audit
GPT 5.5 (xhigh) · 20260517T094700Z
40.4% from 310/767 ProgramBench-scored tests. Raw public eval statuses: failure: 523, passed: 444, skipped: 11.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch.
e9e8360a457ceval.tests.test_oranda_comprehensive.test_build_with_license (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_build_with_description (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_build_with_package_json (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_no_arguments_shows_help (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_build_with_styles (failure)e9e8360a457ceval.tests.test_oranda_comprehensive.test_build_with_repository (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 | 64.9% | $8.72 | 70 |
| 2 | GPT 5.5 (high) | OpenAI | 59.7% | $3.58 | 39 |
| 3 | GPT 5.5 | OpenAI | 55.1% | $1.15 | 23 |
| 4 | Claude Opus 4.7 | Anthropic | 53.6% | $1.36 | 50 |
| 5 | Claude Sonnet 4.6 | Anthropic | 53.6% | $21.15 | 501 |
| 6 | Gemini 3.1 Pro | 43.9% | $0.90 | 80 | |
| 7 | GPT 5.4 | OpenAI | 42.5% | $0.22 | 10 |
| 8 | Gemini 3 Flash | 41.5% | $0.42 | 103 | |
| 9 | Claude Opus 4.6 | Anthropic | 39.6% | $9.62 | 238 |
| 10 | Claude Opus 4.7 (xhigh) | Anthropic | 38.6% | $9.62 | 153 |
| 11 | GPT 5 mini | OpenAI | 31.0% | $0.03 | 19 |
| 12 | GPT 5.4 mini | OpenAI | 19.2% | $0.03 | 9 |
| 13 | Claude Haiku 4.5 | Anthropic | 17.2% | $0.29 | 65 |