PatentAgentBench - Codex - Overall: leaderboard
PatentAgentBench Overall evaluates patent evidence search, comparison, drafting, responses and translation under Codex CLI 0.151.0. Its primary Equal Task Mean Score (%) uses the full assigned case denominator, including final failures after up to three execution-failure retries. Harnesses and unreported model effort remain distinct source configurations. The suite has 510 patent cases; its public banks contain 1,311 records because translation's 80 cases contain 881 segment items.
Metric: Equal Task Mean Score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 (PatentAgentBench Run: 2610.35031v1; GLM-5.3; Codex CLI 0.151.0; effort unreported) | 48.1 | |
| 2 | GPT-5.6 Sol (PatentAgentBench Run: 2610.35031v1; GPT-5.6-Sol; Codex CLI 0.151.0; effort unreported) | 43.8 | |
| 3 | Qwen 3.8 Max (PatentAgentBench Run: 2610.35031v1; Qwen3.8-Max; Codex CLI 0.151.0; effort unreported) | 38.3 | |
| 4 | Kimi K3 (PatentAgentBench Run: 2610.35031v1; Kimi K3; Codex CLI 0.151.0; effort unreported) | 27.1 | |
| 5 | DeepSeek V4 Pro (PatentAgentBench Run: 2610.35031v1; DeepSeek-V4-Pro; Codex CLI 0.151.0; effort unreported) | 26 | |
| 6 | MiniMax-M3 (PatentAgentBench Run: 2610.35031v1; MiniMax-M3; Codex CLI 0.151.0; effort unreported) | 17.5 |
Interactive version: theaggregate.ai/benchmark?slug=patentagentbench-codex-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.