Qwen3.5 122B A10B: Complete Specifications, Pricing, API Access & Use Cases (2026)

Qwen3.5 122B A10B is a large open-weight mixture-of-experts model from the Qwen team at Alibaba, designed for reasoning, coding, tool use, multimodal understanding and agent-oriented workflows. Released in February 2026, it sits between smaller Qwen3.5 deployment options and the much larger 397B-A17B variant. Its combination of 122B total parameters, 10B activated parameters and a native 262K-token context makes it particularly relevant to teams evaluating capable models that can still be self-hosted.

What Is Qwen3.5 122B A10B?

Qwen3.5 122B A10B is a post-trained causal language model with a vision encoder. The official model card documents 122B total parameters with approximately 10B activated, using a sparse mixture-of-experts design with 256 experts, eight routed experts and one shared expert. Qwen combines this MoE structure with Gated Delta Networks and gated attention.

The model was released on February 24, 2026, alongside Qwen3.5-35B-A3B and Qwen3.5-27B. Qwen3.5 itself debuted earlier in February with the larger 397B-A17B model.

Unlike a text-only LLM, this model belongs to Qwen3.5’s unified vision-language architecture. That matters for deployments involving screenshots, documents, charts, images or video frames alongside normal text prompts. It is distributed under the Apache 2.0 license, enabling self-hosted deployment under that license’s terms.

What Are Qwen3.5 122B A10B’s Key Specifications and Pricing?

The official Qwen model card documents 262,144 tokens natively, with extension up to 1,010,000 tokens. Alibaba Cloud separately applies tiered API pricing: for requests up to 128K input tokens it lists $0.115 per million input tokens and $0.917 per million output tokens, while inputs above 128K and up to 256K cost $0.287/M input and $2.294/M output. These are provider-direct rates and should not automatically be treated as Gate.AI billing tiers.

For a simple calculation using the Gate.AI-listed rates, a workload processing 1 million input tokens and producing 250,000 output tokens would be approximately:

$0.115 + (0.25 × $0.917) = $0.34425

before any account-specific conditions or other charges.

What Can Qwen3.5 122B A10B Do That Makes It Useful in Production?

The model is particularly relevant to agentic and multi-step workflows. Qwen’s own evaluations cover tool calling, search agents, planning, coding and general agent benchmarks, while the architecture activates only part of its total parameter pool for each forward pass.

A practical deployment might give the model a long technical document, screenshots from an application and a set of callable tools. It can reason over that combined context, decide which tool is needed and produce a structured textual response. Production systems should still validate tool arguments and consequential actions rather than allowing generated outputs to execute without controls.

Local deployment is another differentiator. Qwen publishes model weights and documents operation with Transformers, vLLM and SGLang. This gives teams with sufficient infrastructure more control over deployment than closed API-only models.

For additional context on larger open-weight alternatives, the Qwen3 235B A22B specifications guide provides a useful comparison point.

What Are Qwen3.5 122B A10B’s Supported Modalities?

Qwen categorizes Qwen3.5 122B A10B as an image-text-to-text model and provides image-plus-text examples directly in its model card. Its official evaluation suite also includes video-understanding benchmarks.

This distinction is important: multimodal understanding does not mean multimodal generation. The verified output is text; the model should not be described as an image, video or audio generator.

Where Does Qwen3.5 122B A10B Fall Short?

The biggest operational constraint is deployment size. Although only 10B parameters are activated for a forward pass, the full model contains 122B parameters, so self-hosting still requires substantial memory and accelerator resources.

Its very long native context also does not mean every application should fill all 262K tokens. Larger prompts increase processing requirements, and Alibaba’s hosted pricing demonstrates that long-context requests may move into a more expensive billing tier.

Finally, some platform-specific details remain separate. A capability documented by Qwen for the open-weight model does not automatically confirm that every Gate.AI endpoint exposes the same modality, parameter or context-extension behavior.

What Is Qwen3.5 122B A10B Best Used For?

Qwen3.5 122B A10B is a strong fit when a workflow needs long-context reasoning, multimodal document understanding, coding, function calling or multi-step agent planning while retaining an open-weight deployment option.

Choose it when 122B-scale model weights are operationally manageable and the extra capability is worth the infrastructure requirement. Consider the smaller Qwen3.5 27B when deployment efficiency matters more, or a hosted model when maintaining large-model inference infrastructure is undesirable.

How Does Qwen3.5 122B A10B Compare to Qwen3.5 35B A3B and Qwen3.5 27B?

All three belong to the same Qwen3.5 generation and support vision-language workflows, but their infrastructure profiles differ substantially. Qwen’s published benchmarks show that no variant dominates every individual evaluation, so model selection should account for workload, hardware, latency targets and cost rather than parameter count alone.

The Qwen3.5 Plus guide is also relevant for teams prioritizing a hosted Qwen experience rather than open-weight deployment.

How Do I Access Qwen3.5 122B A10B Through Gate.AI?

As per the Gate.AI model-card listing, the model identifier is:

alibaba/qwen3.5-122b-a10b

Gate.AI documents an OpenAI-compatible base URL at https://api.gate.ai/openai/v1, Bearer-token authentication and a standard Chat Completions workflow. Gate.AI also advises developers to use an exact model ID obtained from its model catalog, Console or /models endpoint.

The following examples therefore combine Gate.AI’s documented general chat workflow with the model ID shown in the Gate.AI listing. They are documentation-based examples, not execution-confirmed tests.

Python

import os
from openai import OpenAI
client = OpenAI(
    api_key=os.environ["GATEAI_API_KEY"],
    base_url="https://api.gate.ai/openai/v1",
)
try:
    response = client.chat.completions.create(
        model="alibaba/qwen3.5-122b-a10b",
        messages=[
            {
                "role": "user",
                "content": "Create a concise deployment plan for a tool-using AI agent."
            }
        ],
    )
    print(response.choices[0].message.content)
except Exception as error:
    print(f"Gate.AI request failed: {error}")

cURL

curl https://api.gate.ai/openai/v1/chat/completions \
  -H "Authorization: Bearer $GATEAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "alibaba/qwen3.5-122b-a10b",
    "messages": [
      {
        "role": "user",
        "content": "Explain when a mixture-of-experts model is useful."
      }
    ]
  }'

Before production deployment, verify that the model remains available to the account and confirm current rate limits, multimodal request formats and pricing in Gate.AI’s live documentation or Console.

Developers comparing the same gateway workflow across model families may also find the DeepSeek R1 API guide and Llama 3.1 70B guide useful.

FAQs

Is Qwen3.5 122B A10B open weight?

Yes. Qwen publishes the model weights under the Apache 2.0 license and provides documented deployment paths through Transformers, vLLM and SGLang.

How many parameters does Qwen3.5 122B A10B activate?

The official model card specifies 122B total parameters and approximately 10B activated parameters, reflecting its sparse MoE architecture.

Does Qwen3.5 122B A10B support images?

Yes. It accepts image and text inputs and produces text output. Qwen explicitly provides multimodal Transformers and serving examples.

What is its context window?

The native context length is 262,144 tokens. Qwen also documents extension to as much as 1,010,000 tokens for appropriate self-hosted configurations.

How much does Qwen3.5 122B A10B cost on Gate.AI?

As per the Gate.AI listing, the listed rates are $0.115 per million input tokens and $0.917 per million output tokens, with no cache-read or cache-write price listed. Pricing should be rechecked before production use because platform rates can change.