1016generated behavioral tests
85.3%official best score
79.7%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 | 79.7% | ok | 810/1016 | $3.87 | 82 | 0.18h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 77.8% | ok | 790/1016 | n/a | 62 | 0.16h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 76.3% | ok | 775/1016 | n/a | 79 | 0.15h | 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
79.7% from 810/1016 ProgramBench-scored tests. Raw public eval statuses: error: 1, failure: 218, passed: 1093, skipped: 9.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch, terminal rendering mismatch.
dd8712fc730beval.tests.test_batch_comprehensive.test_batch_empty_output (failure)dd8712fc730beval.tests.test_coverage_improvements.test_batch_mode_with_failing_command (failure)dd8712fc730beval.tests.test_complex_scenarios.test_empty_command_output (failure)dd8712fc730beval.tests.test_edge_cases.test_empty_output_consistent (failure)dd8712fc730beval.tests.test_exec_comprehensive.test_exec_with_error_status (failure)dd8712fc730beval.tests.test_error_handling.test_command_with_nonzero_exit (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 (xhigh) | OpenAI | 85.3% | $5.31 | 65 |
| 2 | GPT 5.5 (high) | OpenAI | 82.6% | $4.53 | 59 |
| 3 | Claude Opus 4.6 | Anthropic | 81.1% | $10.71 | 221 |
| 4 | GPT 5.5 | OpenAI | 80.0% | $1.67 | 23 |
| 5 | Claude Opus 4.7 (xhigh) | Anthropic | 76.3% | $10.58 | 210 |
| 6 | Gemini 3.1 Pro | 75.1% | $1.04 | 73 | |
| 7 | Claude Opus 4.7 | Anthropic | 73.5% | $1.23 | 56 |
| 8 | Claude Haiku 4.5 | Anthropic | 66.0% | $0.59 | 88 |
| 9 | Gemini 3 Flash | 61.1% | $0.34 | 71 | |
| 10 | GPT 5.4 | OpenAI | 56.6% | $0.27 | 9 |
| 11 | GPT 5.4 mini | OpenAI | 24.6% | $0.02 | 9 |
| 12 | GPT 5 mini | OpenAI | 2.6% | $0.03 | 10 |
| 13 | Claude Sonnet 4.6 | Anthropic | 1.9% | $14.13 | 378 |