Qwen3.5 397B A17B API pricing

6 providers serve Qwen3.5 397B A17B. Qwen3.5 397B A17B is an open-weight model (Apache-2.0) with 397B total parameters, 17B active per token (mixture-of-experts), up to 262K context. Prices below are per 1M tokens, cheapest input first (self-hosting math further down).

Refreshed 2026-08-28 · list prices from provider APIs

Input floor $0.450/1M at DeepInfra Output floor $3.00/1M at DeepInfra Find your workload winner →

Provider	Input $/1M	Output $/1M	Cache read $/1M	Context
DeepInfra	$0.450	$3.00output floor	$0.220	262K	Use
OpenRouter	$0.600	$3.60	-	262K	Use
Together AI	$0.600	$3.60	$0.350	262K	Use
Scaleway	$0.600	$3.60	-	256K	Use
Tensormesh	$0.600	$3.60	-	262K
Novita AI	$0.600	$3.60	-	262K	Use

The cheapest way to run Qwen3.5 397B A17B

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $122/month, at DeepInfra. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, a busy self-hosted deployment can work out cheaper (~$0.350/1M output on 3× MI300X). See the math below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Workload preset Input tokens / month (millions) Output tokens / month (millions) Input served from cache %

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: Qwen3.5 397B A17B self-hosting economics

Qwen3.5 397B A17B is open-weight (Apache-2.0), so the API price above competes with the GPU-hour market. 397B total parameters (MoE, ~17B active per token) needs roughly 546 GB VRAM at FP8 or 273 GB at INT4, KV-cache headroom included.

No single GPU fits Qwen3.5 397B A17B at FP8 (546 GB needed), so it needs a multi-GPU node. Cheapest tracked option: 3× MI300X (192 GB each) at roughly $1.50/hr total (RunPod). Rent 3× MI300X at RunPod → INT4 quantization drops the requirement to 273 GB. Tune it in the calculator.

Breakeven estimate: a well-batched vLLM deployment on 3× MI300X ($1.50/hr) at ~2382 aggregate tok/s and 50% utilization works out to roughly $0.350 per 1M output tokens, versus $3.00 via the cheapest output-token API (DeepInfra). Self-hosting wins on cost if you can keep the GPU busy. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

Model	Cheapest in $/1M	Cheapest out $/1M	Providers
Qwen3 235B A22B	$0.071	$0.100	10
Qwen3 32B	$0.050	$0.100	9
Qwen3 30B A3B	$0.051	$0.200	8
Qwen3 Next 80B A3B Instruct	$0.090	$0.900	7
Qwen3 Next 80B A3B Thinking	$0.140	$0.900	7
Qwen3-Coder-480b-A35b-Instruct	$0.220	$1.50	7

FAQ

What is the cheapest API for Qwen3.5 397B A17B?

As of 2026-08-28, the lowest input price for Qwen3.5 397B A17B is DeepInfra at $0.450 per 1M input tokens. The lowest output price is DeepInfra at $3.00 per 1M output tokens. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is OpenRouter at $0.600, a 33% difference.

How much VRAM do you need to self-host Qwen3.5 397B A17B?

Qwen3.5 397B A17B has 397B parameters, so plan for roughly 546 GB of VRAM at FP8 or 273 GB at INT4/AWQ, KV-cache headroom included. That exceeds a single GPU. A typical node is 3× MI300X.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON