2933generated behavioral tests
57.1%official best score
55.7%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 | 55.7% | ok | 1634/2933 | $8.19 | 149 | 0.37h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 52.7% | ok | 1547/2933 | n/a | 95 | 0.26h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 35.0% | ok | 1027/2933 | n/a | 104 | 0.24h | 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
55.7% from 1634/2933 ProgramBench-scored tests. Raw public eval statuses: failure: 212, not_run: 1089, passed: 1702, skipped: 2.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, interactive terminal behavior mismatch.
dc4c187c9d8deval.tests.test_errors_only.test_errors_only_schema_validation (failure)dc4c187c9d8deval.tests.test_http_options.test_http_options_with_structured_output (failure)dc4c187c9d8deval.tests.test_output_formats.test_schema_validation_structured_output (failure)dc4c187c9d8deval.tests.test_schema_validation.test_schema_validation_flag_output (failure)dc4c187c9d8deval.tests.test_schema_validation.test_schema_validation_hierarchical_output (failure)dc4c187c9d8deval.tests.test_schema_validation.test_invalid_schema_flag_output (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 | 57.1% | $8.32 | 101 |
| 2 | GPT 5.5 (high) | OpenAI | 55.1% | $2.99 | 37 |
| 3 | Claude Opus 4.6 | Anthropic | 51.7% | $8.76 | 233 |
| 4 | Claude Opus 4.7 (xhigh) | Anthropic | 48.7% | $2.43 | 56 |
| 5 | GPT 5.5 | OpenAI | 47.6% | $1.07 | 20 |
| 6 | Gemini 3 Flash | 40.9% | $0.17 | 80 | |
| 7 | GPT 5.4 mini | OpenAI | 16.8% | $0.05 | 15 |
| 8 | GPT 5 mini | OpenAI | 15.5% | $0.02 | 12 |
| 9 | Claude Opus 4.7 | Anthropic | 0.2% | $2.18 | 61 |
| 10 | Claude Sonnet 4.6 | Anthropic | 0.2% | $15.32 | 406 |
| 11 | GPT 5.4 | OpenAI | 0.2% | $0.18 | 10 |
| 12 | Gemini 3.1 Pro | 0.2% | $0.47 | 33 | |
| 13 | Claude Haiku 4.5 | Anthropic | 0.2% | $0.45 | 113 |