qwen3.8-27b open weights and deployment
Status: Available. Official BF16 safetensors are published under Apache-2.0. The publisher recommends current vLLM, SGLang or TokenSpeed releases for production serving.
Agent deployment prompt
You are my deployment agent. Deploy the official Qwen/Qwen3.8-27B weights as a private, OpenAI-compatible service on this host. Use the publisher's model card as the source of truth: https://huggingface.co/Qwen/Qwen3.8-27B. Preflight - Inspect the OS, GPU model and count, free VRAM, driver and CUDA versions, free disk space, and available ports. Estimate whether the official BF16 weights and the requested context fit with safe headroom for runtime and concurrency. - If the host cannot safely run the official weights, stop before installing or launching anything. Explain the limiting resource and propose a separately reviewed quantized alternative; do not substitute one automatically. Deployment - If preflight passes, create an isolated Python environment, install a current compatible vLLM release, and serve the official model through its OpenAI-compatible API. Generate an API key and store it in a protected environment variable. Bind to 127.0.0.1 first; do not expose the endpoint publicly. - Use the model card's native context by default. Enable the optional 1M YaRN extension only if I explicitly need it, because it may affect shorter inputs. Verification and handoff - Send a real request to /v1/chat/completions and confirm that the model responds. Report the exact launch command, installed versions, local endpoint, test result, measured GPU memory, and log location. Identify any reverse-proxy or firewall changes that still need my approval. Never print the API key or other secrets.
- License
- Apache-2.0
- Hardware note
- The publisher does not state a minimum GPU requirement; capacity depends on precision, context and concurrency, so the Agent recipe performs a hardware preflight before serving.
- Recommended runtime
- vLLM
- Verified date
- 2026-09-14
Published model card
- Parameter count
- 27B
- Context window
- 262,144 native; up to 1,000,000 with YaRN
- Maximum output
- Not stated as a hard maximum; 131,072 final-response tokens recommended for long agentic tasks
- Input/output modalities
- Text, image and video input; text output
Official sources
- Qwen3.8-27B model card — Qwen Team, retrieved 2026-09-14