benchmark evidence
LongBench-v2
LongBench-v2: 503 human-verified long-context tasks (single-doc QA, multi-doc QA, long in-context learning, code repo comprehension) with gold labels and no LLM-judge dependency.
activeupstream source ->
no matched results
No public rows from this source currently map to an active callable model.
what this result means
LongBench-v2: 503 human-verified long-context tasks (single-doc QA, multi-doc QA, long in-context learning, code repo comprehension) with gold labels and no LLM-judge dependency.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on LongBench-v2. Broad task pages require independent corroboration before naming a general winner.
source record
category: long_context
metric: accuracy
matched models: 0
latest source date: date unavailable
direction: higher is better