PluginEval: leaderboard
PluginEval (2026) is a Chinese function-calling benchmark of 3,000 human-verified queries over 54 plugins, with gold calls validated by real API execution and 500 adversarial negatives on which no tool should be called. Query-level accuracy counts a positive query as correct only when every required tool is called with grounded arguments, and a negative query when the model abstains; percent, higher is better.
Metric: Query-level accuracy (%; correct tool calls with correct arguments on positive queries and abstention on negative queries; original difficulty distribution; 3,000 human-verified Chinese function-calling queries over 54 plugins and 63 tools with real API execution (2,500 positive queries requiring grounded calls, 500 adversarial negatives requiring abstention; easy/medium/hard 0.4/0.4/0.2); GPT-5.2 judge anchored to gold annotations). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 68.17 | #102 |
| 2 | Gemini 3.1 Pro (Preview) | 67.78 | #94 |
| 3 | DeepSeek V4 Pro | 64.33 | #188 |
| 4 | Qwen 3 235B A22B | 58.4 | #623 |
| 5 | GPT-5.4 | 52.13 | #123 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=plugineval · How It Works · Data refreshed daily, snapshot 2026-11-13.