# WikiWebFacts ![Category](https://img.shields.io/badge/Category-Knowledge-1f6feb) ![Metric](https://img.shields.io/badge/Metric-%25%20Accuracy-2da44e) ![Language](https://img.shields.io/badge/Language-Russian-8957e5) ![Saturation](https://img.shields.io/badge/Saturation-Low-d29922) > **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](https://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](/models/yandex/aliceai-foundation-80b-a3b-base) | 86.5 | [Yandex model card](https://huggingface.co/yandex/AliceAI-Foundation-80B-A3B-Base) | 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](/benchmarks/yandex/hardmultiqa) | Harder multi-fact questions, 5-shot | Russian | | [TriviaQA](/benchmarks/triviaqa) | Trivia knowledge | English | ## Run WikiWebFacts on Your Model Download the dataset and evaluation protocol from the Hugging Face dataset page. ## Related Models & Benchmarks - [AliceAI-Foundation-80B-A3B-Base](/models/yandex/aliceai-foundation-80b-a3b-base) - [HardMultiQA](/benchmarks/yandex/hardmultiqa) ## Frequently Asked Questions

What is WikiWebFacts?

A 5-shot benchmark from Yandex that measures factual knowledge in Russian.

What is a good WikiWebFacts score?

In Yandex's comparison, 86.5 (AliceAI-Foundation-80B-A3B-Base) leads; strong open base models range from about 62 to 83.

Who created WikiWebFacts?

Yandex, released openly on Hugging Face with the AliceAI-Foundation-80B-A3B-Base model.

WikiWebFacts — Results

WikiWebFacts — Evaluations