Qwen3.5-9B

EXL3 3 bpw · tabbyapi · 64K context · RTX 4060 Ti 8 GB

Alibaba's 9B model from February 2026: small enough for an 8 GB card, with a 64K–128K window. It thinks, calls tools and stays quick on modest hardware.

55 tok/sdecode

927 tok/sprefill

64Kcontext window

Tested on a siblingSep 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 TheMelonGod/Qwen3.5-9B-exl3 \
  --revision 22ef1303062e0f6d0b282440f8c1f685947f4938 \
  --local-dir ~/models/Qwen3.5-9B-exl3-22ef1303

2 Write the server config

cat > qwen3.5-9b.tabbyapi.64k.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: Qwen3.5-9B-exl3-22ef1303
  backend: exllamav3
  inline_model_loading: false
  max_seq_len: 65536
  cache_size: 66560
  cache_mode: Q4
  tensor_parallel: false
  gpu_split_auto: true
  autosplit_reserve: [256]
  chunk_size: 2048
  output_chunking: true
  max_batch_size: 1
  vision: false
  reasoning: true
  start_in_reasoning: auto
  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/Qwen3.5-9B-exl3-22ef1303:/workspace/models/Qwen3.5-9B-exl3-22ef1303:ro \
  -v $PWD/qwen3.5-9b.tabbyapi.64k.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

No RTX 4060 Ti is rentable; this ran on the RTX 4060, the same chip family.

Details

Weights: TheMelonGod/Qwen3.5-9B-exl3 @ 22ef130306 ›
Image: ghcr.io/0xsero/tabbyapi-exl3@sha256:0f83e6198dc3be2561652d8df2525a7d1a69733e12fc8e25cbc5bc319a8a3ad1
Engine profile: tabbyapi-exl3 ›
Recipe file: registry/recipes/nvidia/rtx-4060-ti-8gb/qwen3.5-9b.tabbyapi.64k.json ›
Model released: Feb 26, 2026
Tested: vast, RTX 4060 (sibling: RTX 4060)