NitiBench+ - Gold context 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 March 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5.1 | 8.04 | #10 |
| 2 | GPT-6.1 Sol | 7.69 | #7 |
| 3 | Gemini 4 Argon | 7.63 | #9 |
| 4 | Claude Opus 5.5 | 7.56 | #3 |
| 5 | Muse Spark 1.3 | 7.09 | #31 |
| 6 | Claude Sonnet 5.5 | 6.67 | #14 |
| 7 | DeepSeek V4.1 Flash | 6.38 | #37 |
| 8 | GPT-6.1 Terra | 6.21 | #15 |
| 9 | DeepSeek V4 Flash | 6.19 | #155 |
| 10 | Qwen 3.8 27B | 6.04 | #123 |
| 11 | Claude Haiku 5.5 | 5.36 | #43 |
| 12 | Gemma 4 31B (IT) (Thinking) | 5.16 | #178 (Gemma 4 31B (IT)) |
| 13 | gemma-4-26B-A4B-it (Thinking) | 4.84 | |
| 14 | NVIDIA-Nemotron-3-Super-120B-A12B-BF16 (Thinking) | 4.29 | #394 (NVIDIA-Nemotron-3-Super-120B-A12B-BF16) |
| 15 | Qwen 3 8B | 2.57 | #678 |
No result here: #6 GPT-6 Astra.
Interactive version: theaggregate.ai/benchmark?slug=nitibench-plus-gold-context-in-force · How It Works · Data refreshed daily, snapshot 2026-11-05.