525generated behavioral tests
67.0%official best score
51.4%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) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 51.4% | ok | 270/525 | n/a | 45 | 0.11h | not exported |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 48.0% | ok | 252/525 | n/a | 53 | 0.15h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260517T094700Z |
Codex /goal | Mini-SWE-compatible no internet | 44.4% | ok | 233/525 | $3.55 | 78 | 0.20h | manifest · eval json · eval 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) · 20260518T232006Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260519T215046Z
Public eval evidence has not been exported for this row yet.
GPT 5.5 (xhigh) · 20260517T094700Z
44.4% from 233/525 ProgramBench-scored tests. Raw public eval statuses: error: 3, failure: 337, passed: 493, skipped: 2.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch, terminal rendering mismatch.
aac649a5af1eeval.tests.test_file_operations.test_in_file_nonexistent (failure)aac649a5af1eeval.tests.test_tui_interaction.test_tui_shows_default_text (failure)aac649a5af1eeval.tests.test_tui_interaction.test_tui_launches_and_quits (failure)aac649a5af1eeval.tests.test_tui_interaction.test_tui_backspace (failure)aac649a5af1eeval.tests.test_tui_interaction.test_tui_text_input (failure)aac649a5af1eeval.tests.test_tui_interaction.test_tui_ctrl_u_clear (failure)
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.7 (xhigh) | Anthropic | 67.0% | $12.21 | 249 |
| 2 | GPT 5.5 (xhigh) | OpenAI | 64.4% | $7.60 | 62 |
| 3 | GPT 5.5 (high) | OpenAI | 57.9% | $3.22 | 34 |
| 4 | Claude Opus 4.7 | Anthropic | 57.1% | $3.08 | 108 |
| 5 | Claude Opus 4.6 | Anthropic | 53.5% | $10.10 | 210 |
| 6 | GPT 5.5 | OpenAI | 53.1% | $0.98 | 19 |
| 7 | Gemini 3.1 Pro | 46.1% | $2.81 | 138 | |
| 8 | GPT 5.4 | OpenAI | 33.3% | $0.11 | 11 |
| 9 | GPT 5.4 mini | OpenAI | 31.8% | $0.02 | 7 |
| 10 | Claude Haiku 4.5 | Anthropic | 24.8% | $0.58 | 119 |
| 11 | Gemini 3 Flash | 5.7% | $0.30 | 87 | |
| 12 | Claude Sonnet 4.6 | Anthropic | 2.7% | $22.65 | 354 |
| 13 | GPT 5 mini | OpenAI | 2.7% | $0.01 | 10 |