Multi-Step Planning
Break down complex problems into actionable steps with reliable execution plans.
Qwen3.5-122B-A10B — Large MoE for Advanced Reasoning
Qwen3.5-122B-A10B is a large MoE model for complex reasoning, multi-step planning, and in-depth analysis. Try it free in your browser.
Ready To Chat
Online
ThinkingQwen3.5-122B-A10B is ready
Qwen3.5-122B-A10B is the default model for this page. Large MoE model tuned for harder reasoning, multi-step plans, and detailed answers.
Pick a model, decide whether this needs web search or thinking, then start with a real prompt.
MoE
Analysis
Starter prompts
Free to use on this site via OpenRouter.
Total Params
122B
Active Params
10B
Context
262K native
License
Apache 2.0
Jump to another Qwen model
Overview
The Sweet Spot of the MoE Lineup
Qwen3.5-122B-A10B sits between the compact 35B-A3B and the flagship 397B-A17B. It activates 10B parameters per token — enough for significantly deeper reasoning than 35B-A3B — while keeping inference costs well below the flagship. For many production workloads, this model hits the optimal cost-quality balance.
Deep Expert Routing
10B active parameters per token deliver substantially deeper reasoning than smaller MoE models.
Production Ready
Strong enough for complex tasks while still cost-effective for API-scale deployment.
Long-Form Quality
Maintains coherence and accuracy across extended outputs and multi-turn conversations.
Qwen3.5-122B-A10B Benchmark
How Qwen3.5-122B-A10B compares to nearby models in the Qwen family.
Model
MMLU-Pro
GPQA / GPQA-family
LiveCodeBench v6
Qwen3.5-35B-A3B
Compact MoE model, also the base model behind Qwen3.5-Flash.
Updated 2026-04-02
MMLU-Pro
85.3
GPQA / GPQA-family
84.2
LiveCodeBench v6
74.6
Qwen3.5-122B-A10B
Mid-tier MoE model for deeper reasoning and agent tasks.
Updated 2026-04-02
MMLU-Pro
86.7
GPQA / GPQA-family
86.6
LiveCodeBench v6
78.9
Qwen3.5-397B-A17B
Flagship open-weight Qwen3.5 model, also the base model behind Qwen3.5-Plus.
Updated 2026-04-02
MMLU-Pro
87.8
GPQA / GPQA-family
88.4
LiveCodeBench v6
83.6
Scores are from public model cards and the qwen.ai release page. Hosted models are labeled with their open-weight base.
Updated 2026-04-02Use Cases
What Qwen3.5-122B-A10B Is Best For
This model shines when tasks require sustained reasoning, detailed analysis, or high-quality structured output.
Research & Analysis
Analyze research papers, financial reports, and technical documentation in depth.
Advanced Coding
Handle multi-file refactoring, architecture decisions, and complex debugging.
Long-Form Writing
Produce coherent articles, reports, and documentation over thousands of words.
Data Interpretation
Analyze datasets, explain patterns, and generate insights from structured data.
Agentic Workflows
Power multi-tool agents that need strong reasoning for task orchestration.
FAQ
Qwen3.5-122B-A10B FAQ
Common questions about the large MoE model.
How does 122B-A10B compare to 35B-A3B?
122B-A10B activates over 3x more parameters per token (10B vs 3B) and draws from a much larger expert pool (122B vs 35B). This results in noticeably better reasoning, especially on complex multi-step tasks.
When should I use 397B instead?
Use 397B-A17B when you need the absolute best reasoning quality and are willing to pay higher compute costs. For most production use cases, 122B-A10B provides excellent quality at lower cost.
Can I self-host this model?
Yes, but it requires multi-GPU setups or high-VRAM servers. Quantized versions help reduce requirements. Cloud deployment via vLLM is the most common production setup.
Is it good for coding tasks?
Yes. 122B-A10B handles complex codebases, multi-file reasoning, and architecture-level decisions well — a significant step up from the dense models for programming tasks.
How much VRAM does 122B-A10B need?
Around 40-60 GB at Q4 quantization. Most users run it on multi-GPU setups or cloud instances with 2-4 GPUs.
Is 122B-A10B good for production?
Yes. It offers a strong balance between quality and cost. Many production setups use it as a middle ground between the compact 35B-A3B and the flagship 397B-A17B.
Does 122B-A10B support tool calling?
Yes. All Qwen 3.5 models support function calling. The 122B-A10B size handles multi-step tool chains more reliably than the smaller variants.
What context window does 122B-A10B support?
Qwen3.5-122B-A10B supports 262,144 native tokens and can stretch higher with compatible serving stacks.