Agents' Last Exam v1.1 - Codex Pass Rate: leaderboard
ALE v1.1, 122 tasks. Native harness/model/effort roster without selecting strongest effort. Pass rate counts perfect-score runs; partial-credit score averages graded scores over all 122 tasks, including zero for failed/missing runs. Timeouts retain graded scores. Separate from legacy ALE v1. Codex; Pass Rate. Metric: Pass Rate (%).
Metric: Pass Rate (%). Source: agents-last-exam.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gpt-6-astra (ALE Run: harness=Codex; effort=max; source-model=gpt-6-astra) | 42.62 | |
| 2 | gpt-6-astra (ALE Run: harness=Codex; effort=xhigh; source-model=gpt-6-astra) | 40.16 | |
| 3 | gpt-6-astra (ALE Run: harness=Codex; effort=medium; source-model=gpt-6-astra) | 40.16 | |
| 4 | gpt-6-astra (ALE Run: harness=Codex; effort=high; source-model=gpt-6-astra) | 38.52 | |
| 5 | gpt-6-astra (ALE Run: harness=Codex; effort=low; source-model=gpt-6-astra) | 38.52 | |
| 6 | gpt-5.6-sol (ALE Run: harness=Codex; effort=max; source-model=gpt-5.6-sol) | 30.33 | |
| 7 | minimax-m3-1-flash (ALE Run: harness=Codex; effort=max; source-model=minimax-m3-1-flash) | 29.51 | |
| 8 | gpt-5.6-luna (ALE Run: harness=Codex; effort=none; source-model=gpt-5.6-luna) | 5.74 |
Interactive version: theaggregate.ai/benchmark?slug=agents-last-exam-v1-1-codex-pass-rate · How It Works · Data refreshed daily, snapshot 2026-10-27.