
How to run Qwen3.8 27B
Bedrock, Vertex AI, or your own GPUs · Checked November 21, 2026
Qwen3.8 27B has 27B parameters, so it needs about 65GB of GPU memory at full precision and about 33GB at FP8. A practical starting point is FP8 on 1x 48GB GPU.
You have three ways to run it: a managed host that already serves it, your own GPUs in your cloud, or both behind one endpoint. Then you point your agents at it. Here is what each path takes.
Qwen3.8 27B hardware requirements
| Precision | Memory needed | Smallest setup that fits |
|---|---|---|
| BF16 (full precision) | ~65GB | 1x 80GB GPU (H100 or A100 80GB) |
| FP8 | ~33GB | 1x 48GB GPU (L40S class) |
| 4-bit quantized | ~17GB | 1x 24GB GPU (RTX 4090 class) |
Our estimates: 27B parameters times bytes per parameter, plus 20% for the KV cache and the serving runtime. Using the full 256K-token context needs more. Check the model card for the precision the weights ship in and any recommended settings.
Where to run Qwen3.8 27B
01Amazon Bedrock
Not confirmed as of November 21, 2026We could not confirm Qwen3.8 27B in the Bedrock model catalog. To keep it in AWS anyway, serve it yourself on GPU instances in your account (EC2 or SageMaker) with the setup below, and check the Bedrock catalog again: new open models are added often.
02Google Cloud Vertex AI
Not confirmed as of November 21, 2026We could not confirm Qwen3.8 27B in Vertex AI Model Garden. To keep it in Google Cloud, serve it yourself on GPUs in your project (GKE or Compute Engine) with the setup below.
03Self-host on your own GPUs
vLLM and SGLang support confirmedDownload Qwen/Qwen3.8-27B from Hugging Face and serve it with an inference server that exposes an OpenAI-compatible endpoint. vLLM and SGLang support it. Start with FP8 on 1x 48GB GPU (L40S class), then load-test with your real prompts before you size for production.
- Keep the endpoint private to your VPC, so prompts and payment data never leave your cloud.
- Read the Apache 2.0 license, which allows commercial use.
- Run two replicas or keep a hosted model as a fallback, so agents keep working during an upgrade.
Run Qwen3.8 27B with Runtime
Serving the model is half the job. The other half is giving agents a safe place to use it. Runtime is the AI agent harness for payment and fintech teams, and it is model-neutral: connect Qwen3.8 27B from Bedrock, Vertex AI, or your own endpoint, and your agents use it next to Claude, GPT, or Gemini.
- In your cloud. Runtime runs in your AWS, GCP, or Azure account or fully self-hosted, next to the model.
- Routed by evals. Evals show which steps Qwen3.8 27B handles as well as a frontier model, so routine work moves to the cheaper model.
- Fallbacks. If your endpoint is down or slow, agents fall back to another model instead of stopping.
- Guardrails. Agents start read-only, wait for approval before money moves, and every run is recorded.
Harnesses that can drive Qwen3.8 27B on Runtime:
Running Qwen3.8 27B: FAQ
How much GPU memory does Qwen3.8 27B need?
+
About 65GB at BF16, 33GB at FP8, and 17GB at 4 bits. These are our estimates: the weights plus 20% headroom. Long contexts need more memory for the KV cache.
What is the cheapest way to self-host Qwen3.8 27B?
+
Start with FP8 on 1x 48GB GPU (L40S class). Quantizing further saves memory but can cost accuracy, so test it on your own tasks first.
Is Qwen3.8 27B available on Amazon Bedrock?
+
As of November 21, 2026, we could not confirm Qwen3.8 27B in the Bedrock model catalog. You can still run it in your AWS account on your own GPU instances (EC2 or SageMaker) with vLLM and SGLang.
Is Qwen3.8 27B available on Google Vertex AI?
+
As of November 21, 2026, we could not confirm Qwen3.8 27B in Vertex AI Model Garden. You can still run it in your Google Cloud project on your own GPUs (GKE or Compute Engine) with vLLM and SGLang.
Can I use Qwen3.8 27B with Runtime?
+
Yes. Runtime is model-neutral. Serve Qwen3.8 27B from Bedrock, Vertex AI, or your own endpoint, connect it to Runtime, and agents use it through a harness like OpenCode. Your data stays in your cloud, and evals tell you which steps it can take over from a more expensive model.