Language modelOpen weights

AliceAI Foundation 80B A3B Base

Yandex

Released: September 21, 2026
Data date: October 7, 2026

Input: Text
Output: Text

AliceAI Foundation 80B A3B Base is Yandex's language model, released on September 21, 2026 and trained from scratch, with 80 billion total and 3 billion active parameters. Its open weights use Apache 2.0.

Yandex releases the pretrained Base checkpoint. This variant is identified as a foundation model rather than an instruct or chat model.

Specifications and access

Specification	Value and source
Provider model ID	yandex/AliceAI-Foundation-80B-A3B-BaseSource
Access	Open weights for local deploymentSource
Input modalities	TextSource
Output modalities	TextSource
License	Apache 2.0Source
Checkpoint type	Pretrained base model, not an instruct or chat checkpointSource

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

WikiWebFacts (5-shot) (Show measurement, test conditions, and source)

Source value: 86.5
Score: 86.5
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

HardMultiQA (5-shot) (Show measurement, test conditions, and source)

Source value: 67.9
Score: 67.9
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

CultCat (4-shot) (Show measurement, test conditions, and source)

Source value: 86.5
Score: 86.5
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

TriviaQA (5-shot, LLM-as-a-judge) (Show measurement, test conditions, and source)

Source value: 79
Score: 79
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

EduBench Russian (5-shot) (Show measurement, test conditions, and source)

Source value: 74.2
Score: 74.2
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: education
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

EduBench Literature (5-shot) (Show measurement, test conditions, and source)

Source value: 73.8
Score: 73.8
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: education
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

EduBench History (5-shot) (Show measurement, test conditions, and source)

Source value: 82
Score: 82
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: education
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

EduBench English (5-shot) (Show measurement, test conditions, and source)

Source value: 76.1
Score: 76.1
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: education
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

ExpertFactsQA Medicine (5-shot) (Show measurement, test conditions, and source)

Source value: 63.6
Score: 63.6
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: expert-knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

ExpertFactsQA Law Challenging (5-shot) (Show measurement, test conditions, and source)

Source value: 49.6
Score: 49.6
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: expert-knowledge
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

EGE CoT (5-shot) (Show measurement, test conditions, and source)

Source value: 90.5
Score: 90.5
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: exam
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

MMLU-Pro CoT (5-shot) (Show measurement, test conditions, and source)

Source value: 66.8
Score: 66.8
Metric: reported score; unit not stated by card
Unit: No unit provided
Category: exam
Direction: Higher is better
Source type: vendor-reported
Evaluator: Yandex AliceAI base model card
Status: active
Retrieved at: 2026-10-07
Methodology: Vendor-reported result. Yandex internal evaluation infrastructure, vLLM at temperature 0 for every model. Benchmark-specific shot count, language, CoT and metric are retained in the benchmark name; the card does not state score units or sample counts.
Context: Vendor-reported results from Yandex internal evaluation. Only the AliceAI-Foundation-80B-A3B-Base column is included; no results from later SFT/instruct models are transferred.

Source Yandex AliceAI base model card

Page 1 of 3

Sources and data date

Type	Evidence and data date
Input	Documented modalities. Retrieved October 7, 2026. Source: Yandex release announcement · Yandex release announcement (retrieved October 7, 2026) · Yandex release announcement · Editorial description reviewed October 7, 2026
Output	Documented modalities. Retrieved October 7, 2026. Source: Yandex release announcement · Yandex release announcement (retrieved October 7, 2026) · Yandex release announcement · Editorial description reviewed October 7, 2026
Research status	Research date October 7, 2026. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources	independent-base-evaluation Original wording No independent evaluation of the exact Base checkpoint was verified in this pass. The model card describes these as Yandex internal evaluations.
Additional source	AliceAI Foundation model card (retrieved October 7, 2026) · AliceAI Foundation model card · Editorial description reviewed October 7, 2026
Additional source	Yandex AliceAI base model card (retrieved October 7, 2026)
Additional source	Model metadata (retrieved October 7, 2026)