1338generated behavioral tests
76.2%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 | 1016/1338 | $6.49 | 108 | 0.28h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 73.4% | ok | 982/1338 | n/a | 46 | 0.20h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 64.7% | ok | 866/1338 | n/a | 70 | 0.26h | 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 1016/1338 ProgramBench-scored tests. Raw public eval statuses: failure: 331, passed: 1051, skipped: 2.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch.
e33416c41190eval.tests.test_additional_coverage.test_debug_operations (failure)e33416c41190eval.tests.test_additional_coverage.test_coroutine_operations (failure)e33416c41190eval.tests.test_additional_coverage.test_io_library_operations (failure)e33416c41190eval.tests.test_environment_variables.test_lua_init_with_at_file (failure)e33416c41190eval.tests.test_error_handling.test_division_by_zero (failure)da10ccf83a2ceval.tests.test_coroutines.test_coroutine_yield_values (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 | 76.2% | $5.54 | 63 |
| 2 | GPT 5.5 (high) | OpenAI | 66.1% | $4.34 | 50 |
| 3 | Claude Opus 4.7 | Anthropic | 43.1% | $4.41 | 95 |
| 4 | GPT 5.5 | OpenAI | 43.0% | $1.02 | 14 |
| 5 | Claude Sonnet 4.6 | Anthropic | 34.7% | $138.99 | 993 |
| 6 | GPT 5.4 | OpenAI | 17.8% | $0.20 | 13 |
| 7 | Gemini 3.1 Pro | 13.9% | $0.80 | 64 | |
| 8 | GPT 5 mini | OpenAI | 5.1% | $0.02 | 16 |
| 9 | GPT 5.4 mini | OpenAI | 4.5% | $0.20 | 265 |
| 10 | Gemini 3 Flash | 3.4% | $0.17 | 112 | |
| 11 | Claude Opus 4.7 (xhigh) | Anthropic | 2.2% | $18.58 | 197 |
| 12 | Claude Haiku 4.5 | Anthropic | 2.2% | $1.02 | 157 |
| 13 | Claude Opus 4.6 | Anthropic | 0.0% | $9.84 | 315 |