583generated behavioral tests
91.3%official best score
88.5%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 | 88.5% | ok | 516/583 | $3.71 | 71 | 0.21h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 68.8% | ok | 401/583 | n/a | 41 | 0.12h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 68.4% | ok | 399/583 | n/a | 32 | 0.10h | 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
88.5% from 516/583 ProgramBench-scored tests. Raw public eval statuses: error: 1, failure: 71, not_run: 1, passed: 648, skipped: 1.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch, terminal rendering mismatch.
98d2903ca553eval.tests.test_comprehensive_edge_cases.test_mixed_json_lines (failure)98d2903ca553eval.tests.test_counter_marker.test_backward_compat_marker_counter (failure)98d2903ca553eval.tests.test_with_pty.test_pty_multiple_graphs (failure)98d2903ca553eval.tests.test_with_pty.test_pty_simple (failure)853708b2a69ceval.tests.test_02_environment.test_screen_tmux_rejection (failure)853708b2a69ceval.tests.test_08_rendering_with_fake_tty.test_render_with_fake_tty_single_value (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 | GPT 5.5 (high) | OpenAI | 91.3% | $3.76 | 43 |
| 2 | Claude Opus 4.6 | Anthropic | 89.0% | $8.17 | 139 |
| 3 | GPT 5.5 (xhigh) | OpenAI | 86.6% | $3.95 | 38 |
| 4 | Claude Opus 4.7 (xhigh) | Anthropic | 83.4% | $4.86 | 95 |
| 5 | Claude Opus 4.7 | Anthropic | 82.5% | $2.34 | 85 |
| 6 | GPT 5.5 | OpenAI | 69.0% | $1.90 | 28 |
| 7 | Claude Sonnet 4.6 | Anthropic | 47.7% | $31.83 | 344 |
| 8 | Gemini 3 Flash | 45.1% | $0.15 | 35 | |
| 9 | GPT 5.4 | OpenAI | 43.2% | $0.12 | 11 |
| 10 | Claude Haiku 4.5 | Anthropic | 41.9% | $0.18 | 67 |
| 11 | GPT 5.4 mini | OpenAI | 35.0% | $0.02 | 10 |
| 12 | Gemini 3.1 Pro | 19.9% | $1.63 | 67 | |
| 13 | GPT 5 mini | OpenAI | 16.3% | $0.02 | 16 |