Qwen3.5-122B-A10B (MoE) Architecture Parameters, Benchmarks & VRAM
Sweet-spot 122B MoE (10B active) for 128GB unified-memory workstations (~75GB at Q4_K_M).
🧬 Qwen3.5-122B-A10B (MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (Qwen/Qwen3.5-122B-A10B) config.json & Model Card
1. Backbone & Dimensions
qwen3_5_moe- Total / Active Params
- 122B / 10B
- Hidden Size (d_model)
- 3,072
- MoE Expert Dim
- 1,024
- MoE Experts
- 256 Routed · Top-8 + 1 Shared
- Activation / Norm
- SiLU (SwiGLU) · ε=1e-6
- Native Tensor Dtype
- BF16
2. Attention & Layer Layout
HYBRID- Num Hidden Layers
- 48 Layers (12 Global KV)
- Hidden Layer Layout
- 48 Hybrid MoE Layers: 12 × [3 × Gated DeltaNet→MoE + 1 × Gated Attention→MoE]
- Attention Heads (Q / KV)
- 32 Q / 2 KV (16:1 GQA)
- Head Dimension
- 256
- Linear / KV Compression
- Only 12 layers × 2 KV heads cached; 256 routed experts (8 active)
- Vision Encoder / MTP
- ViT 27L · MTP 1L
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 262,144 tokens
- Max Extended Context
- 1,010,000 tokens
- RoPE Configuration
- mRoPE (θ=10,000,000 · partial=0.25 · YaRN 4x→1M)
- Vocab Size
- 248,320 (Untied)
- Special Token IDs
- BOS: 248044 · EOS: [248046,248044] · PAD: 248044
📋 Qwen3.5-122B-A10B (MoE) Official API & System Card Specifications
Verified against official Alibaba Qwen developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- Qwen/Qwen3.5-122B-A10B
- Release Date
- 2026-02-23
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 256,000 tokens (256K)
- Max Output Tokens
- 66,000 tokens
- Input Modalities
- Text + Image + Video
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 0.8
- Cached Input / 1M
- ¥ 0.8
- Output / 1M Tokens
- ¥ 6.38
- Tiered Pricing Notes
- SiliconFlow: Open-weight cloud hosting on SiliconFlow
🏆 Official Benchmark Results
Sourced from official System Card / Model Card & technical reports
A more robust and challenging multi-task language understanding benchmark that extends MMLU by expanding multiple-choice options from 4 to 10, eliminating trivial questions, and focusing on reasoning-intensive tasks.
C-Eval is a comprehensive Chinese evaluation suite designed to assess advanced knowledge and reasoning abilities of foundation models in a Chinese context.
PhD-level Google-proof science reasoning questions
A verified subset of 500 software engineering problems from real GitHub issues, validated by human annotators for evaluating language models' ability to resolve real-world coding issues by generating patches for Python codebases.
LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code.
Berkeley Function Calling Leaderboard V4 (BFCL-V4) evaluates LLMs on their ability to accurately call functions and APIs, including simple, multiple, parallel, and nested function calls across diverse programming scenarios.
MMMU (Massive Multi-discipline Multimodal Understanding) is a benchmark designed to evaluate multimodal models on college-level subject knowledge and deliberate reasoning.
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 Qwen3.5-122B-A10B (MoE) Full Benchmark Scorecard
81 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 4 items
A verified subset of 500 software engineering problems from real GitHub issues, validated by human annotators for evaluating language models' ability to resolve real-world coding issues by generating patches for Python codebases.
English subset of FullStackBench for evaluating end-to-end software engineering and full-stack development capability.
Chinese subset of FullStackBench for evaluating end-to-end software engineering and full-stack development capability.
Terminal-Bench 2.0 is an updated benchmark for testing AI agents' tool use ability to operate a computer via terminal.
🤖 Agents & Tools 11 items
t2-bench is a benchmark for evaluating agentic tool use capabilities, measuring how well models can select, sequence, and utilize tools to solve complex tasks. It tests autonomous planning and execution in multi-step scenarios.
Berkeley Function Calling Leaderboard V4 (BFCL-V4) evaluates LLMs on their ability to accurately call functions and APIs, including simple, multiple, parallel, and nested function calls across diverse programming scenarios.
A high-difficulty benchmark purpose-built to comprehensively evaluate LLM agents on the Chinese web, consisting of 289 multi-hop questions spanning 11 diverse domains including Film & TV, Technology, Medicine, and History.
AndroidWorld Success Rate (SR) benchmark - A dynamic benchmarking environment for autonomous agents operating on Android devices.
BrowseComp is a benchmark comprising 1,266 questions that challenge AI agents to persistently navigate the internet in search of hard-to-find, entangled information.
WideSearch is an agentic search benchmark that evaluates models' ability to perform broad, parallel search operations across multiple sources. It tests wide-coverage information retrieval and synthesis capabilities.
OSWorld-Verified is a verified subset of OSWorld, a scalable real computer environment for multimodal agents supporting task setup, execution-based evaluation, and interactive learning across Ubuntu, Windows, and macOS.
A tool-calling and multimodal interaction benchmark for testing visual instruction following and execution reliability.
Seal-0 is a benchmark for evaluating agentic search capabilities, testing models' ability to navigate and retrieve information using tools.
VITA-Bench evaluates AI agents on real-world virtual task automation, measuring their ability to complete complex multi-step tasks in simulated environments.
DeepPlanning evaluates LLMs on complex multi-step planning tasks requiring long-horizon reasoning, goal decomposition, and strategic decision-making.
📐 Math 13 items
An improved version of the MMLU benchmark featuring manually re-annotated questions to identify and correct errors in the original dataset. Provides more reliable evaluation metrics for language models by addressing dataset quality issues found in the original MMLU.
Harvard-MIT Mathematics Tournament 2025 - A prestigious student-organized mathematics competition for high school students featuring two tournaments (November 2025 at MIT and February 2026 at Harvard) with individual tests, team rounds, and guts rounds.
Harvard-MIT Mathematics Tournament 2025 - A prestigious student-organized mathematics competition for high school students featuring two tournaments (November 2025 at MIT and February 2026 at Harvard) with individual tests, team rounds, and guts rounds.
MathVista-Mini is a smaller version of the MathVista benchmark that evaluates mathematical reasoning in visual contexts.
A more robust and challenging multi-task language understanding benchmark that extends MMLU by expanding multiple-choice options from 4 to 10, eliminating trivial questions, and focusing on reasoning-intensive tasks.
Multilingual Massive Multitask Language Understanding dataset released by OpenAI, featuring professionally translated MMLU test questions across 14 languages including Arabic, Bengali, German, Spanish, French, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Swahili…
MATH-Vision is a dataset designed to measure multimodal mathematical reasoning capabilities.
A multimodal mathematics and reasoning benchmark focused on dynamic visual problem solving.
Extended version of MMLU-Pro providing additional challenging multiple-choice questions for evaluating language models across diverse academic and professional domains. Built on the foundation of the Massive Multitask Language Understanding benchmark framework.
Polymath is a challenging multi-modal mathematical reasoning benchmark designed to evaluate the general cognitive reasoning abilities of Multi-modal Large Language Models (MLLMs).
SuperGPQA is a comprehensive benchmark that evaluates large language models across 285 graduate-level academic disciplines.
Humanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions.
A competitive programming benchmark using problems from the CodeForces platform. The benchmark evaluates code generation capabilities of LLMs on algorithmic problems with difficulty ratings ranging from 800 to 2400.
👁️ Multimodal 39 items
A vision-language benchmark that probes blind spots and brittle reasoning in multimodal models.
AI2D is a dataset of 4,903 illustrative diagrams from grade school natural sciences (such as food webs, human physiology, and life cycles) with over 15,000 multiple choice questions and answers.
A visual reasoning benchmark evaluating multimodal inference under challenging spatial and grounded tasks.
Version 1.1 of MMBench, an improved bilingual benchmark for assessing multi-modal capabilities of vision-language models through multiple-choice questions in both English and Chinese, providing systematic evaluation across diverse vision-language tasks.
OCRBench: Comprehensive evaluation benchmark for assessing Optical Character Recognition (OCR) capabilities in Large Multimodal Models across text recognition, scene text VQA, and document understanding tasks.
OmniDocBench 1.5 is a comprehensive benchmark for evaluating multimodal large language models on document understanding tasks, including OCR, document parsing, information extraction, and visual question answering across diverse document types.
The first-ever comprehensive evaluation benchmark of Multi-modal LLMs in Video analysis. Features 900 videos (254 hours) with 2,700 question-answer pairs covering 6 primary visual domains and 30 subfields.
A comprehensive benchmark for multi-task long video understanding that evaluates multimodal large language models on videos ranging from 3 minutes to 2 hours across 9 distinct tasks including reasoning, captioning, recognition, and summarization.
RealWorldQA is a benchmark designed to evaluate basic real-world spatial understanding capabilities of multimodal models.
Video-MME is a comprehensive evaluation benchmark for multi-modal large language models in video analysis. It features 900 videos across 6 primary visual domains with 30 subfields, ranging from 11 seconds to 1 hour in duration, with 2,700 question-answer pairs.
MMMU (Massive Multi-discipline Multimodal Understanding) is a benchmark designed to evaluate multimodal models on college-level subject knowledge and deliberate reasoning.
MMStar is an elite vision-indispensable multimodal benchmark comprising 1,500 challenge samples meticulously selected by humans to evaluate 6 core capabilities and 18 detailed axes.
Video-MMMU evaluates Large Multimodal Models' ability to acquire knowledge from expert-level professional videos across six disciplines through three cognitive stages: perception, comprehension, and adaptation.
A comprehensive OCR benchmark for evaluating Large Multimodal Models (LMMs) in literacy. Comprises four OCR-centric tracks: multi-scene text reading, multilingual text reading, document parsing, and key information extraction.
A semantically-labeled knowledge-enhanced dataset for medical visual question answering. Contains 642 radiology images (CT scans, MRI scans, X-rays) covering five body parts and 14,028 bilingual English-Chinese question-answer pairs annotated by experienced physicians.
A benchmark for multimodal spatial-language understanding and visual-linguistic question answering.
CharXiv-R is the reasoning component of the CharXiv benchmark, focusing on complex reasoning questions that require synthesizing information across visual chart elements.
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings.
A comprehensive multi-modal video understanding benchmark covering 20 challenging video tasks that require temporal understanding beyond single-frame analysis.
MMVU (Multimodal Multi-disciplinary Video Understanding) is a benchmark for evaluating multimodal models on video understanding tasks across multiple disciplines, testing comprehension and reasoning capabilities on video content.
LVBench is an extreme long video understanding benchmark designed to evaluate multimodal models on videos up to two hours in duration. It contains 6 major categories and 21 subcategories, with videos averaging five times longer than existing datasets.
ScreenSpot-Pro is a novel GUI grounding benchmark designed to rigorously evaluate the grounding capabilities of multimodal large language models (MLLMs) in professional high-resolution computing environments.
A comprehensive benchmark designed to evaluate image-context reasoning in large visual-language models (LVLMs) by challenging models with 346 images and 1,129 carefully crafted questions to assess language hallucination and visual illusion.
A comprehensive benchmark to evaluate expert-level medical knowledge and advanced reasoning, featuring 4,460 questions spanning 17 specialties and 11 body systems.
A medical visual question answering benchmark built on biomedical literature and medical figures.
Embodied Reasoning Question Answering benchmark consisting of 400 multiple-choice visual questions across spatial reasoning, trajectory reasoning, action reasoning, state estimation, and multi-view reasoning for evaluating AI capabilities in physical world interactions.
Object Detection in the Wild (ODinW) benchmark for evaluating object detection models' task-level transfer ability across diverse real-world datasets in terms of prediction accuracy and adaptation efficiency.
A benchmark for early-stage visual reasoning and perception on child-like vision tasks.
A multimodal benchmark for scene understanding and reasoning over the nuScenes autonomous driving domain.
CountBench evaluates object counting capabilities in visual understanding.
RefCOCO-avg measures object grounding accuracy averaged across RefCOCO, RefCOCO+, and RefCOCOg benchmarks.
EmbSpatialBench evaluates embodied spatial understanding and reasoning capabilities.
RefSpatialBench evaluates spatial reference understanding and grounding.
SimpleVQA is a visual question answering benchmark focused on simple queries.
MMLongBench-Doc evaluates long document understanding capabilities in vision-language models.
SUNRGBD evaluates RGB-D scene understanding and 3D grounding capabilities.
ZEROBench-Sub is a subset of the ZEROBench benchmark.
Hypersim evaluates 3D grounding and depth understanding in synthetic indoor scenes.
ZEROBench is a challenging vision benchmark designed to test models on zero-shot visual understanding tasks.
📜 Long Context 2 items
🧠 Reasoning & Science 6 items
C-Eval is a comprehensive Chinese evaluation suite designed to assess advanced knowledge and reasoning abilities of foundation models in a Chinese context.
Global PIQA is a multilingual commonsense reasoning benchmark that evaluates physical interaction knowledge across 100 languages and cultures.
A challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy.
LiveCodeBench is a holistic and contamination-free evaluation benchmark for large language models for code.
MultiChallenge is a realistic multi-turn conversation evaluation benchmark that challenges frontier LLMs across four key categories: instruction retention (maintaining instructions throughout conversations), inference memory (recalling and connecting details from previous…
OJBench is a competition-level code benchmark designed to assess the competitive-level code reasoning abilities of large language models. It comprises 232 programming competition problems from NOI and ICPC, categorized into Easy, Medium, and Hard difficulty levels.
📚 Knowledge & General 6 items
Instruction-Following Evaluation (IFEval) benchmark for large language models, focusing on verifiable instructions with 25 types of instructions and around 500 prompts containing one or more verifiable constraints.
MAXIFE is a multilingual benchmark evaluating LLMs on instruction following and execution across multiple languages and cultural contexts.
INCLUDE is a multilingual language-understanding benchmark of 197,243 multiple-choice questions drawn from regional exams in 44 languages, built to measure regional and cultural knowledge rather than translated English content (Romanou et al., 2024).
WMT24++ is a comprehensive multilingual machine translation benchmark that expands the WMT24 dataset to cover 55 languages and dialects.
Instruction Following Benchmark evaluating model's ability to follow complex instructions.
NOVA-63 is a multilingual evaluation benchmark covering 63 languages, designed to assess LLM performance across diverse linguistic contexts and tasks.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ Qwen3.5-122B-A10B (MoE) Interactive VRAM & Token Speed Calculator
Physical 128 GB · Load 71%
MoE: reads only 6.06 GB active weights/step
Enable Q8/Q4 KV quantization to scale batch concurrency
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3.5:122b-a10bMac Studio M5 Max 128G Fits Qwen3.5-122B-A10B (MoE) · Cost & TCO
Approx. $3,499 (~$27/GB VRAM) · 120W TDP
📊 Qwen3.5-122B-A10B (MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 244 GB | 0.52 GB | 250.7 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 129.63 GB | 0.52 GB | 134.04 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 100.04 GB | 0.52 GB | 103.86 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 86.77 GB | 0.52 GB | 90.33 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 73.96 GB | 0.52 GB | 77.26 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 64.81 GB | 0.52 GB | 67.93 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 59.63 GB | 0.52 GB | 62.64 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 48.04 GB | 0.52 GB | 50.82 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for Qwen3.5-122B-A10B (MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of Qwen3.5-122B-A10B (MoE)?
Qwen3.5-122B-A10B (MoE) uses Qwen3_5MoeForConditionalGeneration with 48 layers (48 Hybrid MoE Layers: 12 × [3 × Gated DeltaNet→MoE + 1 × Gated Attention→MoE]), hidden_size=3072, 32 Q heads / 2 KV heads (head_dim=256), and vocab_size=248,320.
How much VRAM does Qwen3.5-122B-A10B (MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), Qwen3.5-122B-A10B (MoE) requires 77.26 GB VRAM. Full FP16 requires 250.7 GB.