Qwen3.5-9B

reasoning
multimodal

Qwen3.5-9B is a compact dense model with 9 billion parameters, retaining the key capabilities of the flagship versions in the series. Built on 32 layers, it utilizes the same hybrid architecture as the larger models: interleaving blocks with Gated DeltaNet and Gated Attention (in a 3:1 ratio), ensuring a balance between speed, accuracy, and memory efficiency on long contexts. This allows it to natively process up to 262k tokens effectively, confidently competing with models 2–3 times its size. The model supports native multimodality with understanding of text, images, and video, making it a versatile solution for various tasks.

The model's results are very impressive for its "weight class." In knowledge tests (MMLU-Pro – 82.5), it surpasses many larger models, including GPT-OSS-120B. In instruction following (IFEval – 91.5), it shows results close to top-tier models. Its agentic capabilities particularly stand out: TAU2-Bench (79.1) and BFCL-V4 (66.1) — results that were only achievable by models of the 70B+ scale just a year ago. Its multimodal capabilities are also top-notch: MathVision (78.9), MMMU-Pro (70.1), OCRBench (89.2), and VlmsAreBlind (93.7) demonstrate a deep understanding of visual information.

The model is optimal for scenarios where a balance between performance and resources is crucial. It requires only 8GB of RAM to run in a quantized format, making it accessible for consumer hardware. It is ideal for building OCR systems, real-time document and image analysis, and tasks requiring fast data processing locally within an enterprise perimeter.


Announce Date: 27.02.2026
Parameters: 10B
Context: 263K
Layers: 32, using full attention: 8
Attention Type: Linear Attention
Developer: Qwen
Transformers Version: 4.57.0.dev0
vLLM Version: 0.17.0
License: Apache 2.0

Public endpoint

Use our pre-built public endpoints for free to test inference and explore Qwen3.5-9B capabilities. You can obtain an API access token on the token management page after registration and verification.
Model Name Context Type GPU Status Link
There are no public endpoints for this model yet.

Private server

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

  • maximize endpoint performance,
  • enable full context for long sequences,
  • ensure top-tier security for data processing in an isolated, dedicated environment,
  • use custom weights, such as fine-tuned models or LoRA adapters.

Recommended server configurations for hosting Qwen3.5-9B

Prices:
Name GPU Price, hour TPS Max Concurrency
teslaa10-1.16.32.160
262,144.0
1 $0.53 1.336 Launch
teslat4-2.16.32.160
262,144.0
tensor
2 $0.54 1.920 Launch
teslaa2-2.16.32.160
262,144.0
tensor
2 $0.57 1.920 Launch
rtx3090-1.16.24.160
262,144.0
1 $0.83 1.336 Launch
rtx2080ti-3.12.24.120
262,144.0
pipeline
3 $0.84 1.721 Launch
rtx4090-1.16.32.160
262,144.0
1 $1.02 1.336 Launch
rtx2080ti-4.16.32.160
262,144.0
tensor
4 $1.12 2.640 Launch
teslav100-1.12.64.160
262,144.0
1 $1.20 2.230 Launch
rtxa5000-2.16.64.160.nvlink
262,144.0
tensor
2 $1.23 3.709 Launch
rtx3080-3.16.64.160
262,144.0
pipeline
3 $1.43 1.385 Launch
rtx5090-1.16.64.160
262,144.0
1 $1.59 2.230 Launch
rtx3080-4.16.64.160
262,144.0
tensor
4 $1.82 2.193 Launch
teslaa100-1.16.64.160
262,144.0
1 $2.37 7.598 Launch
h100-1.16.64.160
262,144.0
1 $3.83 7.598 Launch
h100nvl-1.16.96.160
262,144.0
1 $4.11 9.163 Launch
h200-1.16.128.160
262,144.0
1 $4.74 14.419 Launch
Prices:
Name GPU Price, hour TPS Max Concurrency
teslaa10-1.16.32.160
262,144.0
1 $0.53 0.963 Launch
teslat4-2.16.32.160
262,144.0
tensor
2 $0.54 1.547 Launch
teslaa2-2.16.32.160
262,144.0
tensor
2 $0.57 1.547 Launch
rtx3090-1.16.24.160
262,144.0
1 $0.83 0.963 Launch
rtx2080ti-3.12.24.120
262,144.0
pipeline
3 $0.84 1.348 Launch
rtx4090-1.16.32.160
262,144.0
1 $1.02 0.963 Launch
rtx2080ti-4.16.32.160
262,144.0
tensor
4 $1.12 2.268 Launch
teslav100-1.12.64.160
262,144.0
1 $1.20 1.858 Launch
rtxa5000-2.16.64.160.nvlink
262,144.0
tensor
2 $1.23 3.336 Launch
rtx3080-3.16.64.160
262,144.0
pipeline
3 $1.43 1.013 Launch
rtx5090-1.16.64.160
262,144.0
1 $1.59 1.858 Launch
rtx3080-4.16.64.160
262,144.0
tensor
4 $1.82 1.820 Launch
teslaa100-1.16.64.160
262,144.0
1 $2.37 7.225 Launch
h100-1.16.64.160
262,144.0
1 $3.83 7.225 Launch
h100nvl-1.16.96.160
262,144.0
1 $4.11 8.791 Launch
h200-1.16.128.160
262,144.0
1 $4.74 14.047 Launch
Prices:
Name GPU Price, hour TPS Max Concurrency
teslat4-3.32.64.160
262,144.0
pipeline
3 $0.88 2.202 Launch
teslaa10-2.16.64.160
262,144.0
tensor
2 $0.93 2.512 Launch
teslat4-4.16.64.160
262,144.0
tensor
4 $0.96 3.680 Launch
teslaa2-3.32.128.160
262,144.0
pipeline
3 $1.06 2.202 Launch
rtx2080ti-4.16.32.160
262,144.0
tensor
4 $1.12 1.444 Launch
teslav100-1.12.64.160
262,144.0
1 $1.20 1.034 Launch
rtxa5000-2.16.64.160.nvlink
262,144.0
tensor
2 $1.23 2.512 Launch
teslaa2-4.32.128.160
262,144.0
tensor
4 $1.26 3.680 Launch
rtx3090-2.16.64.160
262,144.0
tensor
2 $1.56 2.512 Launch
rtx5090-1.16.64.160
262,144.0
1 $1.59 1.034 Launch
rtx3080-4.16.64.160
262,144.0
tensor
4 $1.82 0.996 Launch
rtx4090-2.16.64.160
262,144.0
tensor
2 $1.92 2.512 Launch
teslaa100-1.16.64.160
262,144.0
1 $2.37 6.401 Launch
h100-1.16.64.160
262,144.0
1 $3.83 6.401 Launch
h100nvl-1.16.96.160
262,144.0
1 $4.11 7.967 Launch
h200-1.16.128.160
262,144.0
1 $4.74 13.223 Launch

Related models

Need help?

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.