Skip to main content

Qwen3 Coder 30B A3B (AWQ)

Large MoE code model — 30B total, ~3B active parameters per token, AWQ 4-bit weights.

Overview​

  • model_name: hf-Qwen-Qwen3-Coder-30B-A3B-Instruct-FP8 (HuggingFace import, currently free). The FP8 in the identifier is historical: the served weights are the AWQ quantization below.
  • Source: QuantTrio/Qwen3-Coder-30B-A3B-Instruct-AWQ
  • Scale: 30B total / ~3B active (Mixture-of-Experts) — big-model quality at small-model speed
  • Served on: GPU via vLLM. Placement follows free capacity, so the board it runs on — and with it the context window — can change from one load to the next. Read max_model_len from /v1/models instead of assuming a number.
  • Status: ready (verified end-to-end through the public inference route, 2026-08-02)

When to use​

✅ Code generation, refactoring, repository-level reasoning, and agentic coding tasks where you want top-tier code quality above the 7B class. The MoE design keeps per-token cost close to a small dense model.

❌ For quick low-latency completions, Qwen 2.5 Coder 7B answers faster and cheaper.

Live specs​

Context window, VRAM footprint, and readiness come from the live catalog:

curl "https://api.colabhive.com/api/builder/v1/endpoints?visibility=public&search=Qwen3-Coder-30B" \
  -H "X-API-Key: $COLABHIVE_API_KEY"

Quick start​

curl -X POST "https://api.colabhive.com/api/builder/v1/endpoints/<endpoint_id>/infer" \
  -H "X-API-Key: $COLABHIVE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"input": {"messages": [{"role": "user", "content": "Write a Python function that merges two sorted lists in O(n)."}], "max_tokens": 512}}'