CIPHER Cross-Record Inference - FIR (Chain-of-Thought): leaderboard
Metric: Harmonic answer score (%; per question the harmonic mean of token overlap with the reference and Gemini 2.5 Flash semantic acceptance, zero when rejected, averaged; First Information Reports (Indian police records) with typed placeholders from post-hoc masking; the full support-controlled context of 39-42 records including distractors; chain-of-thought prompting; greedy decoding). Source: arxiv.org. Saturation forecast: Around January 2027. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.0 Flash | 43.5 |
| 2 | GPT-OSS-20B | 34.7 |
| 3 | Qwen 3 Next 80B A3B | 34.3 |
| 4 | GPT-OSS-120B | 34 |
| 5 | Llama 3.3 70B Instruct | 31 |
Interactive version: theaggregate.ai/benchmark?slug=cipher-cross-record-inference-fir-chain-of-thought · How It Works · Data refreshed daily, snapshot 2026-09-26.