MicrosoftLanguage model

WizardLM-2 8x22B

WizardLM-2 8x22B has no overall index value yet because there are too few independent benchmark results.

PopularityRank 400 of 785

36out of 100-1 vs. last month

based on search, usage and attention

Summary

WizardLM-2 8x22B is Microsoft’s large sparse Mixture-of-Experts instruction model. The checkpoint has 141 billion total parameters, with roughly 39 billion activated per token. Microsoft positions it as the largest WizardLM-2 variant for complex conversation, multilingual work, reasoning, and agentic tasks. Its context window is 65,536 tokens.

The model card documents serving through Transformers, vLLM, and SGLang. Its model list also includes quantized versions for llama.cpp, Ollama, and LM Studio. The Apache-2.0 license applies to the published checkpoint.

See for yourself: one prompt, one attempt

SVG short test

One SVG task with the same fixed settings for every model, without hidden thinking.

WizardLM-2 8x22B

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 62 published benchmark results

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.

IFEval strict accuracy (Show measurement, test conditions, and source)

Source value: 52.72
Score: 52.72
Metric: 0-shot strict accuracy
Unit: %
Benchmark version: Open LLM Leaderboard row; IFEval 0-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; IFEval is 0-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

BBH normalized accuracy (Show measurement, test conditions, and source)

Source value: 48.58
Score: 48.58
Metric: 3-shot accuracy
Unit: %
Benchmark version: Open LLM Leaderboard row; BBH 3-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; BBH is 3-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

MATH Lvl 5 exact match (Show measurement, test conditions, and source)

Source value: 22.28
Score: 22.28
Metric: 4-shot exact match
Unit: %
Benchmark version: Open LLM Leaderboard row; MATH Level 5 4-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; MATH Level 5 is 4-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

GPQA accuracy (Show measurement, test conditions, and source)

Source value: 17.56
Score: 17.56
Metric: 0-shot accuracy
Unit: %
Benchmark version: Open LLM Leaderboard row; GPQA 0-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; GPQA is 0-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

MuSR accuracy (Show measurement, test conditions, and source)

Source value: 14.54
Score: 14.54
Metric: 0-shot accuracy
Unit: %
Benchmark version: Open LLM Leaderboard row; MuSR 0-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; MuSR is 0-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

MMLU-Pro accuracy (Show measurement, test conditions, and source)

Source value: 39.96
Score: 39.96
Metric: 5-shot accuracy
Unit: %
Benchmark version: Open LLM Leaderboard row; MMLU-Pro 5-shot
Category: general
Direction: Higher is better
Source type: official-leaderboard
Evaluator: WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table
Status: active
Retrieved at: 2026-10-15
Methodology: Official benchmark leaderboard. Exact alpindale/WizardLM-2-8x22B model-index value; MMLU-Pro is 5-shot as specified in the model card.
Context: The model card embeds Open LLM Leaderboard values but does not pin the evaluated repository revision or complete decoding configuration.

Source WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table

AlpacaEval 2.0 length-controlled win rate (Show measurement, test conditions, and source)

Source value: 51.3
Score: 51.3
Metric: LC win rate
Unit: %
Benchmark version: AlpacaEval 2.0
Category: general
Direction: Higher is better
Source type: independent-evaluator
Evaluator: OSC: Cognitive Orchestration through Dynamic Knowledge Alignment in Multi-Agent LLM Collaboration, Table 1
Status: active
Retrieved at: 2026-10-17
Methodology: Independent evaluation. OSC paper Table 1. The authors explicitly say the WizardLM-2-8x22B results are from prior work. AlpacaEval 2.0 uses 805 instructions compared with GPT-4 Preview and a GPT-4-based evaluator; MT-Bench uses published multi-turn dialogue scores.
Context: Prior-work result reproduced by the paper, not a new run by its authors. Scored N, confidence intervals, and complete generation configuration are not given.

Source OSC: Cognitive Orchestration through Dynamic Knowledge Alignment in Multi-Agent LLM Collaboration, Table 1

AlpacaEval 2.0 standard win rate (Show measurement, test conditions, and source)

Source value: 62.3
Score: 62.3
Metric: standard win rate
Unit: %
Benchmark version: AlpacaEval 2.0
Category: general
Direction: Higher is better
Source type: independent-evaluator
Evaluator: OSC: Cognitive Orchestration through Dynamic Knowledge Alignment in Multi-Agent LLM Collaboration, Table 1
Status: active
Retrieved at: 2026-10-17
Methodology: Independent evaluation. OSC paper Table 1. The authors explicitly say the WizardLM-2-8x22B results are from prior work. AlpacaEval 2.0 uses 805 instructions compared with GPT-4 Preview and a GPT-4-based evaluator; MT-Bench uses published multi-turn dialogue scores.
Context: Prior-work result reproduced by the paper, not a new run by its authors. Scored N, confidence intervals, and complete generation configuration are not given.

Source OSC: Cognitive Orchestration through Dynamic Knowledge Alignment in Multi-Agent LLM Collaboration, Table 1

DynaCode (Show measurement, test conditions, and source)

Source value: 35.4
Score: 35.4
Metric: Pass@1; Unit 1
Unit: %
Benchmark version: Findings of ACL 2025 paper, Table 2; DynaCode
Direction: Higher is better
Source type: independent-evaluator
Evaluator: DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Exact paper row WizardLM-2-8x22B (176B); Pass@1 means percentage of problems solved correctly on the first attempt, with temperature 0. The paper says results are averaged over three experiments using five random seeds. Row-level scored denominator, served checkpoint revision, and complete generation settings are not bound to this row.
Context: The paper prints parenthetical ± values for Table 2; their meaning is not mapped to a confidence interval here. The table's Average is retained as the publisher-labeled value, not recalculated.

Source DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133

DynaCode (Show measurement, test conditions, and source)

Source value: 27.5
Score: 27.5
Metric: Pass@1; Unit 2
Unit: %
Benchmark version: Findings of ACL 2025 paper, Table 2; DynaCode
Direction: Higher is better
Source type: independent-evaluator
Evaluator: DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Exact paper row WizardLM-2-8x22B (176B); Pass@1 means percentage of problems solved correctly on the first attempt, with temperature 0. The paper says results are averaged over three experiments using five random seeds. Row-level scored denominator, served checkpoint revision, and complete generation settings are not bound to this row.
Context: The paper prints parenthetical ± values for Table 2; their meaning is not mapped to a confidence interval here. The table's Average is retained as the publisher-labeled value, not recalculated.

Source DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133

DynaCode (Show measurement, test conditions, and source)

Source value: 25.1
Score: 25.1
Metric: Pass@1; Unit 3
Unit: %
Benchmark version: Findings of ACL 2025 paper, Table 2; DynaCode
Direction: Higher is better
Source type: independent-evaluator
Evaluator: DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Exact paper row WizardLM-2-8x22B (176B); Pass@1 means percentage of problems solved correctly on the first attempt, with temperature 0. The paper says results are averaged over three experiments using five random seeds. Row-level scored denominator, served checkpoint revision, and complete generation settings are not bound to this row.
Context: The paper prints parenthetical ± values for Table 2; their meaning is not mapped to a confidence interval here. The table's Average is retained as the publisher-labeled value, not recalculated.

Source DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133

DynaCode (Show measurement, test conditions, and source)

Source value: 12.8
Score: 12.8
Metric: Pass@1; Unit 4
Unit: %
Benchmark version: Findings of ACL 2025 paper, Table 2; DynaCode
Direction: Higher is better
Source type: independent-evaluator
Evaluator: DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Exact paper row WizardLM-2-8x22B (176B); Pass@1 means percentage of problems solved correctly on the first attempt, with temperature 0. The paper says results are averaged over three experiments using five random seeds. Row-level scored denominator, served checkpoint revision, and complete generation settings are not bound to this row.
Context: The paper prints parenthetical ± values for Table 2; their meaning is not mapped to a confidence interval here. The table's Average is retained as the publisher-labeled value, not recalculated.

Source DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133

Page 1 of 6

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

The 100 most popular models of this category, 0 to 100, higher is more popular

20 of 101 models

GPT-6 Astra83

Claude Haiku 5.577

GPT-6 Sol73

GPT-573

DeepSeek V4.1 Flash 101769

Kimi K367

MiniCPM-V 4.7 35B-A3B63

WizardLM-2 8x22B36

DeepSeek V4.1 Flash 1017

MiniCPM-V 4.7 35B-A3B

GPT-6.1 Sol

Claude Opus 5

Qwen 3.8 Flash

WizardLM-2 8x22B

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 7, 2026

WizardLM-2 8x22BAion 3.5 36Claude Opus 4.6 36ZDTaichu5.0-9B 36Apodex 1.1 Mini 36

Method changes in the timeline, a jump at these points is no market event:

1November 1, 2026: Value only with search or usage data, 30-day attention

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

Similar models

More models of the same kind as WizardLM-2 8x22B, chosen by model line, developer, size, price and release date.

All specifications

Architecture

Parameters: 141 billion; Source: Parameters
Active parameters: 39 billion; Source: Active parameters

Input and output

Context window: 65,536 tokens; Source: Context window
Inputs: Text; Source: Inputs
Outputs: Text; Source: Outputs

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking: Unknown; no source
Watermarking technique: Not known; no source

Licence and access

Weights: Open weights; Source: Weights
Access: API; Source: Access
License: Apache 2.0; Source: License

Frequently asked questions about WizardLM-2 8x22B

How large is the context window of WizardLM-2 8x22B?

65,536 tokens. That is how much text WizardLM-2 8x22B reads at once, about 100 A4 pages.

Page count as a rule of thumb from Artificial Analysis

Context windowTokens the model reads at once · higher is better

GLM-5.1200K

DeepSeek-V3 0324164K

GLM-4.5131K

GPT OSS 120B131K

DeepSeek-V3.2128K

DeepSeek-R1128K

WizardLM-2 8x22BRank 103 of 10366K

Can I run WizardLM-2 8x22B locally?

Yes, the weights are open. With 141B parameters it needs a lot of memory, more a server than a laptop.

Stated by Microsoft

Open weights: yes
License: Apache 2.0
Parameters: 141B
active per token: 39B

Sources and data date

Show all 18 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

Type	Evidence and data date
Input	Documented modalities. Source: Hugging Face · Hugging Face · WizardLM-2 8x22B model card · Editorial description reviewed October 2, 2026
Output	Documented modalities. Source: Hugging Face · Hugging Face · WizardLM-2 8x22B model card · Editorial description reviewed October 2, 2026
Research status	Research date October 15, 2026. 10 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not published	checkpoint-revision Note The embedded leaderboard table does not pin the exact weight revision and some task/date details.
Not published	scored-denominator-and-generation-settings Note The cited result tables omit task-level scored denominators and complete sampling/decoding details. The OSC paper marks the WizardLM results as prior work.
Not published	score-unit Original wording The inspected Table 3 excerpt presents numeric scores without an explicit unit label; no percent unit is inferred.
Not published	confidence-interval-95 Original wording The paper does not report a 95% confidence interval for these cells.
Not published	immutable-api-served-revision Original wording The paper identifies Lepton AI API as the route but no immutable served revision.
Additional source	WizardLM-2 8x22B model-card Open LLM Leaderboard evaluation table (retrieved October 15, 2026)
Additional source	TIGER-Lab official MMLU-Pro leaderboard results.csv (retrieved October 15, 2026)
Additional source	OSC: Cognitive Orchestration through Dynamic Knowledge Alignment in Multi-Agent LLM Collaboration, Table 1 (retrieved October 17, 2026)
Additional source	Arena Learning: Build Data Flywheel for LLMs, Table 8 (retrieved October 17, 2026)
Additional source	DARG (NeurIPS 2024) (retrieved October 18, 2026)
Additional source	Finance Language Model Evaluation (FLAME) (retrieved October 19, 2026)
Additional source	DynaCode paper, Findings of ACL 2025, ACL Anthology ID 2025.findings-acl.1133 (retrieved October 19, 2026)
Additional source	NeurIPS 2024 Datasets and Benchmarks Track paper publishing REPLIQA0 results (retrieved October 19, 2026)
Additional source	Model metadata (retrieved October 15, 2026)
Additional source	Model metadata (retrieved October 17, 2026)
Additional source	WizardLM-2 official release post · Editorial description reviewed October 2, 2026
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Ollama: cumulative downloads for the popularity index
Additional source	Replicate: cumulative runs for the popularity index
Additional source	Google Trending Now: limited feed selection for Germany and the US
Additional source	Creative Commons Attribution 4.0