586generated behavioral tests
91.8%official best score
85.2%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 | 85.2% | ok | 499/586 | $5.54 | 93 | 0.31h | manifest · eval json · eval summary · usage audit |
| 2 | GPT 5.5 (xhigh) | 20260519T215046Z |
Codex /goal | Paper prompt + Goal contract no internet | 80.0% | ok | 469/586 | n/a | 36 | 0.15h | not exported |
| 3 | GPT 5.5 (xhigh) | 20260518T232006Z |
Codex /goal | Paper prompt + /goal no internet | 78.7% | ok | 461/586 | n/a | 28 | 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
85.2% from 499/586 ProgramBench-scored tests. Raw public eval statuses: failure: 96, passed: 588, skipped: 1.
Agent trace summary unavailable.
Likely miss class: behavioral mismatch, exact version/output mismatch, interactive terminal behavior mismatch.
aa2ac39b58a8tests.test_basic_invocation.test_help_flag (failure)aa2ac39b58a8tests.test_environment_variables.test_invalid_restart_signal (failure)aa2ac39b58a8tests.test_environment_variables.test_kill_signal_rejected (failure)aa2ac39b58a8tests.test_status_formatting.test_status_script_incompatible_with_restart (failure)aa2ac39b58a8tests.test_environment_variables.test_empty_restart_signal (failure)bcc0b83fd418tests.test_basic_invocation.test_help_flag_shows_summary (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 | 91.8% | $6.62 | 83 |
| 2 | GPT 5.5 (high) | OpenAI | 88.9% | $1.64 | 23 |
| 3 | Claude Sonnet 4.6 | Anthropic | 88.6% | $12.14 | 383 |
| 4 | Claude Opus 4.6 | Anthropic | 80.9% | $7.88 | 202 |
| 5 | Claude Opus 4.7 | Anthropic | 78.5% | $1.89 | 77 |
| 6 | GPT 5.5 | OpenAI | 75.8% | $0.86 | 19 |
| 7 | Gemini 3.1 Pro | 70.8% | $1.98 | 152 | |
| 8 | Gemini 3 Flash | 65.7% | $0.25 | 77 | |
| 9 | GPT 5.4 | OpenAI | 59.2% | $0.24 | 13 |
| 10 | Claude Opus 4.7 (xhigh) | Anthropic | 42.8% | $33.60 | 454 |
| 11 | Claude Haiku 4.5 | Anthropic | 35.5% | $0.53 | 109 |
| 12 | GPT 5.4 mini | OpenAI | 13.3% | $0.02 | 7 |
| 13 | GPT 5 mini | OpenAI | 1.2% | $0.01 | 12 |