Qwen3.8-27B

EXL3 6 bpw · tabbyapi · 256K context · RTX PRO 6000 Blackwell 96 GB

Alibaba's 27B dense model from August 2026, and the strongest model that fits one consumer card. It thinks before it answers, calls tools reliably and reads images, which makes it the default for coding agents and long projects.

52 tok/sdecode

568 tok/sprefill

256Kcontext window

Tested on this cardSep 25, 2026

Run it

With Omarchy Local AI it is one button. By hand, it is three steps: the weights, the config, the container. The server then answers on http://localhost:8000/v1.

1 Download the weights

hf download turboderp/Qwen3.8-27B-exl3 \
  --revision 60d005a257b39ecb25e4ba23c1dd29df877d5c69 \
  --local-dir ~/models/Qwen3.8-27B-exl3-60d005a2

2 Write the server config

cat > qwen3.8-27b.tabbyapi.256k.yml <<'EOF'
network: {host: 0.0.0.0, port: 5000, disable_auth: true, disable_fetch_requests: true, send_tracebacks: false, api_servers: [OAI], sse_ping_interval: 15}
logging: {log_prompt: false, log_generation_params: false, log_requests: false, log_chat_completion_requests: false}
model:
  model_dir: /workspace/models
  model_name: Qwen3.8-27B-exl3-60d005a2
  backend: exllamav3
  inline_model_loading: false
  max_seq_len: 262144
  cache_size: 1052672
  cache_mode: Q8
  tensor_parallel: false
  gpu_split_auto: true
  autosplit_reserve: [256]
  chunk_size: 2048
  output_chunking: true
  max_batch_size: 4
  vision: true
  reasoning: true
  start_in_reasoning: always
  tool_format: qwen3_coder
  template_vars_default: {}
draft_model: {draft_mode: mtp}
memory: {sysmem_recurrent_cache: 4096, sysmem_kv_cache: 0, cuda_malloc_async: true}
EOF

3 Start the server

docker run --rm \
  --gpus all \
  -p 8000:5000 \
  --shm-size 8g \
  -v ~/models/Qwen3.8-27B-exl3-60d005a2:/workspace/models/Qwen3.8-27B-exl3-60d005a2:ro \
  -v $PWD/qwen3.8-27b.tabbyapi.256k.yml:/app/config.yml:ro \
  --entrypoint /opt/venv/bin/python3 \
  ghcr.io/0xsero/tabbyapi-exl3@sha256:0f83e6198dc3be2561652d8df2525a7d1a69733e12fc8e25cbc5bc319a8a3ad1 \
  main.py \
  --config /app/config.yml

What it passed

✓ loadLoads and serves within an hour

✓ chatAnswers a plain question and stops on its own

✓ reasoningThinks separately and gets 17 × 23 right

✓ toolsCalls a tool with the right arguments and uses the result

✓ contextRecalls a code buried in a prompt that fills 85% of the window

✓ speedDecodes at 15 tok/s or more

Passed all six checks on a real RTX PRO 6000 WS on Sep 25, 2026.

Details

Weights: turboderp/Qwen3.8-27B-exl3 @ 60d005a257 ›
Image: ghcr.io/0xsero/tabbyapi-exl3@sha256:0f83e6198dc3be2561652d8df2525a7d1a69733e12fc8e25cbc5bc319a8a3ad1
Engine profile: tabbyapi-exl3 ›
Recipe file: registry/recipes/nvidia/rtx-pro-6000-blackwell-96gb/qwen3.8-27b.tabbyapi.256k.json ›
Model released: Aug 5, 2026
Tested: vast, RTX PRO 6000 WS