NitiBench+ - RAG In-force: leaderboard
NitiBench+ Table 17:500 held-out Thai legal questions; three distinct evidence settings. Bar-standard answer scores 0–10; citation precision/recall/F 1 fractions 0–1. avg@3 for 13 self-hosted bf 16 models; other rows are leaderboard API observations, not replicated avg@3. AppendixC exact checkpoints and reasoning-on controls retained; other API effort unspecified. Same official judging rubric; citations checked against current in-force law. The public live site was 403, resolved via complete native paper table. Metric: Bar-standard score (0–10).
Metric: Bar-standard score (0–10). Source: arxiv.org. Saturation forecast: Around February 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5.1 | 7.78 | #10 |
| 2 | Gemini 4 Argon | 7.54 | #9 |
| 3 | GPT-6.1 Sol | 7.28 | #7 |
| 4 | Claude Opus 5.5 | 7.03 | #3 |
| 5 | DeepSeek V4.1 Flash | 6.39 | #37 |
| 6 | Muse Spark 1.3 | 6.37 | #31 |
| 7 | GPT-6.1 Terra | 6.14 | #15 |
| 8 | Claude Sonnet 5.5 | 5.97 | #14 |
| 9 | DeepSeek V4 Flash | 5.71 | #155 |
| 10 | Qwen 3.8 27B | 5.48 | #123 |
| 11 | Claude Haiku 5.5 | 4.62 | #43 |
| 12 | Gemma 4 31B (IT) (Thinking) | 4.57 | #178 (Gemma 4 31B (IT)) |
| 13 | gemma-4-26B-A4B-it (Thinking) | 3.68 | |
| 14 | NVIDIA-Nemotron-3-Super-120B-A12B-BF16 (Thinking) | 3.23 | #394 (NVIDIA-Nemotron-3-Super-120B-A12B-BF16) |
| 15 | Qwen 3 8B | 0.96 | #678 |
No result here: #6 GPT-6 Astra.
Interactive version: theaggregate.ai/benchmark?slug=nitibench-plus-rag-in-force · How It Works · Data refreshed daily, snapshot 2026-11-05.