What hardware does Qwen3.8 27B run on?

Tuned builds: NVIDIA H100 SXM5. plowrt also runs them on H100 NVL, H200 SXM.

Does Qwen3.8 27B fit on one GPU?

Yes. The sm90a-h100-tp1 build runs on one NVIDIA H100 SXM5.

Can I run Qwen3.8 27B on AMD?

Not with a tuned build yet. Today's builds are for NVIDIA.

What context length does Qwen3.8 27B support?

The upstream model supports 262,144 tokens. Our builds serve up to 16,384.

What license is Qwen3.8 27B under?

Apache-2.0. Check the checkpoint's license on Hugging Face before commercial use.

How do I run Qwen3.8 27B?

Install plowrt, then run plowrt load qwen3.8-27b --fetch-weights and plowrt serve --model qwen3.8-27b --port 8080. It serves an OpenAI-compatible API.

LLM 27B 256k native context bf16 Apache-2.0

Qwen3.8 27B

Qwen3.8-27B with gated DeltaNet recurrent layers, in bf16 and FP8 on NVIDIA.

Qwen/Qwen3.8-27B

Specs

Developer: Qwen team, Alibaba
Parameters: 27B
Task: LLM
Context: 262,144 tokens upstream, 16,384 served by our builds
Precisions: bf16
License: Apache-2.0
Checkpoint: Qwen/Qwen3.8-27B
Updated: 29 Oct 2026

Measured A specific workload and configuration against vLLM.

On one H100 in BF16 against vLLM 0.28, with 1,000 input tokens and 128 output tokens: sampled decoding at temperature 1 and top-p 0.95 reaches 0.85 to 0.96 times vLLM’s total throughput. P99 time to first token is 0.48 times vLLM’s at one request, but 1.70 times at 64 requests. Two repeats per cell.

Run of 13 Oct 2026. Settings and every row. More results on benchmarks.

Where it runs Each build is tuned for one part. plowrt runs it beyond that.

Tuned for

NVIDIA H100 SXM5

Runs as-is on H100 NVL, H200 SXM

Elsewhere plowrt still runs the model on other NVIDIA, AMD and Apple hardware, without this build’s tuning: expect lower throughput.

Run it An OpenAI-compatible server on your own machine.

plowrt 0.1.3 Install, pull, serve, call

curl -fsSL https://pdist.infervisor.ai/install.sh | sh

Install

```
plowrt load qwen3.8-27b --fetch-weights
```
Pull with weights

plowrt serve --model qwen3.8-27b --port 8080

Serve

curl -s http://127.0.0.1:8080/v1/chat/completions \
  -H 'content-type: application/json' \
  -d '{"model":"qwen3.8-27b","messages":[{"role":"user","content":"Hello"}]}'

Try it

Builds `plowrt load` picks the closest fit for your machine.

NVIDIA H100 SXM5 80 GiB, also H100 NVL, H200 SXM bf16 1 GPU 16k Published

plowrt load infervisor/qwen3.8-27b:sm90a-h100-tp1 --fetch-weights

This build only

Target: sm90a-h100-tp1 , tuned on NVIDIA H100 SXM5; runs as-is on H100 NVL, H200 SXM; other hardware runs it untuned
Checkpoint: Qwen/Qwen3.8-27B
Published: 29 Oct 2026
Compiled by: plowc 0.1.3 , from this recipe

About

The FP8 label is the canonical production recipe for this model on H100: W8A8 weights, recurrent GDN state kept on device, and a decode ladder of 1 to 48 slots. The bf16 label serves the unmodified checkpoint.

Questions

What hardware does Qwen3.8 27B run on?: Tuned builds: NVIDIA H100 SXM5. plowrt also runs them on H100 NVL, H200 SXM.
Does Qwen3.8 27B fit on one GPU?: Yes. The sm90a-h100-tp1 build runs on one NVIDIA H100 SXM5.
Can I run Qwen3.8 27B on AMD?: Not with a tuned build yet. Today's builds are for NVIDIA.
What context length does Qwen3.8 27B support?: The upstream model supports 262,144 tokens. Our builds serve up to 16,384.
What license is Qwen3.8 27B under?: Apache-2.0. Check the checkpoint's license on Hugging Face before commercial use.
How do I run Qwen3.8 27B?: Install plowrt, then run plowrt load qwen3.8-27b --fetch-weights and plowrt serve --model qwen3.8-27b --port 8080. It serves an OpenAI-compatible API.