TIDE (Evolving Documents, Gold Context) - Context Misalignment: leaderboard
Part of TIDE (Sobhani et al., 2026) on temporally evolving Bangladesh customs instruments, with the governing instrument supplied in the prompt: a misleading preamble states a superseded version and the model must answer from the version actually in force, over 116 questions judged by a three-model council. Strict accuracy in percent, higher is better.
Metric: Strict accuracy (%; 116 questions whose preamble states a superseded version, judged by the council against the value in force; TIDE, 3,050 expert-verified questions over 644 Bangladesh customs instruments (1969-2025); free-form answers are judged by a council of Claude Opus 4.6, Gemini 3.5 Flash and GPT-5.1 by majority vote with a hard date gate, deterministic tasks by exact match; strict accuracy without the partial-rubric credit; thinking enabled, temperature 0; gold context: the governing instrument text is supplied in the prompt). Source: arxiv.org. Saturation forecast: Around 2029. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 (Thinking) | 27.62 | #256 (Kimi K2.5) |
| 2 | GPT-5.2 (Thinking) | 26.73 | #176 (GPT-5.2) |
| 3 | Qwen 3.6 35B A3B | 22.42 | #462 |
| 4 | GPT-OSS-120B | 19.89 | #700 |
| 5 | Gemini 3.5 Flash | 19.88 | #101 |
| 6 | Gemma 4 31B (Reasoning) | 19.85 | #394 (Gemma 4 31B) |
| 7 | GLM-4.6 | 17.23 | #486 |
| 8 | Claude Sonnet 4.5 (Thinking) | 14.71 | #250 (Claude Sonnet 4.5) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=tide-evolving-documents-gold-context-context-misalignment · How It Works · Data refreshed daily, snapshot 2026-11-13.