benchmark evidence
AlpacaEval 2.0 LC Win Rate
Instruction following evaluated via length-controlled win rate.
activeupstream source ->
winner on AlpacaEval 2.0 LC Win Rate
direct benchmark result, not a broad vertical composite | source row dated 2026-05-08
scored on 2026-05-08
latest mapped results | top 20
| # | Model | Score | Evidence | Tested |
|---|---|---|---|---|
| 1 | Meta: Llama 3.3 70B Instruct | 34.4 | model-only independent_benchmark | 2026-05-08 |
what this result means
Instruction following evaluated via length-controlled win rate.
This benchmark contributes direct public evidence. Read its scope before generalizing the result.
A win here is a win on AlpacaEval 2.0 LC Win Rate. Broad task pages require independent corroboration before naming a general winner.
source record
category: writing
metric: lc_win_rate
matched models: 1
latest source date: 2026-05-08
direction: higher is better