LoCoEval (Multi-Hop) - Information Item Extraction: leaderboard
Metric: F1 (0-1, times 100) on the 160 information-item extraction tasks of the multi-hop subset (64 query outlines built on DevEval repositories): list every requirement detail the conversation gave for a target function, matched to the ground truth by a Gemini 2.5 Flash judge; the Full setting: the standalone model reads the whole long-horizon repository-development conversation (30 to 70 turns, 64K to 256K tokens, generated by a Gemini 2.5 Flash mock user from LoCoEval's query outlines) truncated only at its context window, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 235B A22B 2507 Instruct | 42.1 | #291 |
| 2 | DeepSeek V3.2 | 41.6 | #198 |
| 3 | GPT-5 Mini | 38.5 | #176 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=locoeval-multi-hop-information-item-extraction · How It Works · Data refreshed daily, snapshot 2026-10-11.