WikiWebFacts — Results
| Rank | Model | Score |
|---|---|---|
| 1 | aliceai-foundation-80b-a3b-base | 86.5 |
WikiWebFacts
1 phaseActive
Yandex's 5-shot benchmark of Russian-language factual knowledge, released openly alongside AliceAI-Foundation-80B-A3B-Base.
Overview
WikiWebFacts
Quick answer: WikiWebFacts is a 5-shot benchmark of factual knowledge in Russian, published by Yandex together with its evaluation protocol and AliceAI-Foundation-80B-A3B-Base. The model scores 86.5, ahead of DeepSeek-V4-Flash-Base (83.2), Nemotron-3-Super-120B-A12B-Base (72.8) and GLM-4.5-Air-Base (70.2) in Yandex's own comparison.
At a Glance
What it tests: Recall of facts about entities in a Russian-language context, evaluated 5-shot on base models.
Why it matters: Most factual-knowledge benchmarks are English-centric; WikiWebFacts measures how much Russian-language world knowledge a model has absorbed during pre-training.
Known limitations: Published scores come from Yandex's own evaluation infrastructure (vLLM, temperature 0) and compare base models only.
What WikiWebFacts Measures
WikiWebFacts probes factual knowledge that appears on Russian Wikipedia and the Russian web. It is run 5-shot, so it suits base (pre-trained, not instruction-tuned) models, and is reported together with HardMultiQA, a harder multi-hop sibling released at the same time.
Benchmark Specifications
| Field | Value |
|---|---|
| Task category | Knowledge (Russian) |
| Metric | % accuracy, 5-shot |
| Created by | Yandex |
| Dataset | huggingface.co/datasets/yandex/WikiWebFacts |
How WikiWebFacts Is Scored
Models are prompted with five examples and scored on factual accuracy using Yandex's published evaluation protocol.
State-of-the-Art Results
| Rank | Model | Score | Source | Date |
|---|---|---|---|---|
| 1 | AliceAI-Foundation-80B-A3B-Base | 86.5 | Yandex model card | 2026-09 |
Yandex's own comparison (not on Benchgen): DeepSeek-V4-Flash-Base 83.2, Nemotron-3-Super-120B-A12B-Base 72.8, GLM-4.5-Air-Base 70.2, Qwen3.5-35B-A3B-Base 62.4.
WikiWebFacts on Benchgen
No Benchgen results yet — be the first to run WikiWebFacts.
WikiWebFacts vs Other Benchmarks
| Benchmark | What it tests | Language |
|---|---|---|
| WikiWebFacts | Factual knowledge, 5-shot | Russian |
| HardMultiQA | Harder multi-fact questions, 5-shot | Russian |
| TriviaQA | Trivia knowledge | English |
Run WikiWebFacts on Your Model
Download the dataset and evaluation protocol from the Hugging Face dataset page.