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 | 91.8% | ok | 690/752 | $4.60 | 90 | 0.22h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 80.3% | ok | 604/752 | $2.81 | 50 | 0.18h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 52.3% | ok | 393/752 | $3.27 | 70 | 0.21h | 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
91.8% from 690/752 ProgramBench-scored tests. Raw public eval statuses: failure: 63, passed: 783.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch.
c4f46363268eeval.tests.test_basic_invocation.test_invalid_command (failure)c4f46363268eeval.tests.test_basic_invocation.test_no_arguments_shows_help (failure)c4f46363268eeval.tests.test_error_conditions.test_enter_vault_with_no_args (failure)c4f46363268eeval.tests.test_error_conditions.test_invalid_item_type (failure)61cbf5f33b3ctests.test_config_operations.test_invalid_config_type_rejected (failure)61cbf5f33b3ctests.test_config_operations.test_set_conflict_invalid_boolean_panics (failure)
GPT 5.5 (xhigh) · 20260519T215046Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260518T232006Z
52.3% from 393/752 ProgramBench-scored tests. Raw public eval statuses: failure: 363, passed: 483.
Agent trace: 145 shell command(s), 107 executable-observation command(s), 4 build command(s), 2 blocked attempt(s). Raw Codex logs are not published.
Likely miss class: behavioral mismatch.
c4f46363268eeval.tests.test_edge_cases.test_move_vault_to_same_location (failure)c4f46363268eeval.tests.test_edge_cases.test_list_empty_vault (failure)c4f46363268eeval.tests.test_edge_cases.test_create_folder_with_special_chars (failure)c4f46363268eeval.tests.test_edge_cases.test_note_name_with_unicode (failure)c4f46363268eeval.tests.test_edge_cases.test_vault_name_with_spaces (failure)c4f46363268eeval.tests.test_edge_cases.test_create_note_with_special_chars (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 | 97.1% | $9.55 | 73 |
| 2 | GPT 5.5 (high) | OpenAI | 94.1% | $3.55 | 46 |
| 3 | GPT 5.5 | OpenAI | 86.4% | $0.94 | 15 |
| 4 | Claude Opus 4.6 | Anthropic | 84.6% | $8.70 | 216 |
| 5 | Claude Opus 4.7 (xhigh) | Anthropic | 79.4% | $27.97 | 362 |
| 6 | Gemini 3.1 Pro | 76.2% | $2.34 | 222 | |
| 7 | Claude Opus 4.7 | Anthropic | 71.3% | $6.83 | 182 |
| 8 | Claude Sonnet 4.6 | Anthropic | 68.4% | $9.27 | 374 |
| 9 | GPT 5.4 | OpenAI | 63.0% | $0.52 | 14 |
| 10 | Gemini 3 Flash | 41.5% | $0.45 | 166 | |
| 11 | Claude Haiku 4.5 | Anthropic | 37.8% | $1.91 | 245 |
| 12 | GPT 5.4 mini | OpenAI | 7.4% | $0.04 | 11 |
| 13 | GPT 5 mini | OpenAI | 3.9% | $0.03 | 17 |