419generated behavioral tests
78.5%official best score
75.9%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 | 75.9% | ok | 318/419 | $4.75 | 82 | 0.20h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 54.2% | ok | 227/419 | n/a | 56 | 0.16h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 39.4% | ok | 165/419 | n/a | 69 | 0.20h | 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
75.9% from 318/419 ProgramBench-scored tests. Raw public eval statuses: error: 1, failure: 158, passed: 771, skipped: 9.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch, terminal rendering mismatch.
c3c602c9a96feval.tests.test_comprehensive_messages_v2.test_print_app_state_and_quit_msg (failure)c3c602c9a96feval.tests.test_comprehensive_messages_v2.test_change_directory_msg (failure)c3c602c9a96feval.tests.test_message_errors.test_message_with_invalid_field (failure)c3c602c9a96feval.tests.test_message_handling.test_print_msg_in_invalid_yaml (failure)c3c602c9a96feval.tests.test_path_handling.test_stdin_invalid_path (failure)c3c602c9a96feval.tests.test_tmux_tui_interactions.test_tmux_basic_navigation_down (failure)
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.
Official ProgramBench Results by Model
| # | Model | Provider | Score | Cost | Calls |
|---|---|---|---|---|---|
| 1 | GPT 5.5 (xhigh) | OpenAI | 78.5% | $9.61 | 90 |
| 2 | Claude Opus 4.6 | Anthropic | 75.4% | $16.61 | 277 |
| 3 | GPT 5.5 (high) | OpenAI | 68.3% | $3.51 | 37 |
| 4 | Claude Opus 4.7 | Anthropic | 60.5% | $4.36 | 108 |
| 5 | Claude Opus 4.7 (xhigh) | Anthropic | 58.8% | $2.74 | 65 |
| 6 | Gemini 3.1 Pro | 43.0% | $1.76 | 87 | |
| 7 | GPT 5.4 | OpenAI | 41.1% | $0.26 | 11 |
| 8 | Claude Haiku 4.5 | Anthropic | 40.1% | $0.61 | 123 |
| 9 | GPT 5.5 | OpenAI | 36.8% | $1.47 | 22 |
| 10 | GPT 5.4 mini | OpenAI | 36.0% | $0.02 | 7 |
| 11 | Gemini 3 Flash | 35.2% | $0.21 | 65 | |
| 12 | Claude Sonnet 4.6 | Anthropic | 13.8% | $19.67 | 407 |
| 13 | GPT 5 mini | OpenAI | 13.6% | $0.02 | 12 |