Dedicated model · Available as managed deployment
A mixture-of-experts Qwen with roughly 3B active parameters per token, validated on AxForge hardware and deployed on a dedicated DGX Spark for your traffic only. Available as a managed deployment: an OpenAI-compatible endpoint on your own dedicated machine.
Specifications
| Model | Qwen3.6 35B A3B, mixture-of-experts, latest Qwen generation |
|---|---|
| Served model name | qwen3.6-35b-a3b-nvfp4 |
| Architecture | MoE: 35B total, roughly 3B active parameters per token |
| Context window | 65,536 tokens |
| Quantization | NVFP4 |
| Availability | Available as managed deployment, operated by AxForge on a dedicated NVIDIA DGX Spark |
| Hardware | NVIDIA DGX Spark (GB10, 128 GB unified memory), owned and operated by AxForge |
| Region | Málaga, Spain (eu-es-1) |
| Pricing | Hardware from €0.85/hour, rental term: Hour, week, month or year. Managed service quoted per deployment. Request deployment |
Performance
| Metric | Value | Condition |
|---|---|---|
| Decode speed (community) | 86.3 tokens/s | SparkBench PBM @ 4k context, single GB10, NVFP4 + MTP |
Community measurement from SparkBench. MoE with ~3B active parameters per token is a natural fit for GB10. We haven't published our own tuned numbers yet. More on the GB10 page; for quality benchmarks, see the official model card.
Deployment
| System | A dedicated NVIDIA DGX Spark (GB10, 128 GB unified memory), reserved for you |
|---|---|
| Commercial model | Hour, week, month or year. Hardware from €0.85/hour, managed service quoted per deployment |
| Endpoint | OpenAI-compatible /v1 on your own machine, only your traffic |
| Location | Hosted in the EU: Málaga, Spain (eu-es-1) |
| Data handling | Zero prompt retention, same policy as the serverless API |
The model is validated on AxForge hardware and deployed on your system. After deployment, your endpoint speaks the OpenAI API:
# On your dedicated endpoint, after deployment. Its address and key come with it.
import os
from openai import OpenAI
client = OpenAI(base_url="https://YOUR_ENDPOINT/v1", api_key=os.environ["AXFORGE_API_KEY"])
r = client.chat.completions.create(
model="qwen3.6-35b-a3b-nvfp4",
messages=[{"role": "user", "content": "Hello from your own machine"}],
)
print(r.choices[0].message.content)// On your dedicated endpoint, after deployment. Its address and key come with it.
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://YOUR_ENDPOINT/v1", apiKey: process.env.AXFORGE_API_KEY });
const r = await client.chat.completions.create({
model: "qwen3.6-35b-a3b-nvfp4",
messages: [{ role: "user", content: "Hello from your own machine" }],
});
console.log(r.choices[0].message.content);Data & privacy
Prompts and completions are processed in memory in the EU, not written to disk, not logged, not retained, and never used to train anything. We keep only request metadata (token counts, timestamps, status) for billing and operations. The full policy is at axforge.ai/privacy.
FAQ
Not on the serverless API. It is available as a managed deployment: validated on AxForge hardware and deployed on a dedicated DGX Spark for your traffic only. The serverless API serves Qwen3.8 27B.
Qwen3.6 35B A3B is a mixture-of-experts model: 35B total parameters, roughly 3B active per token. That is why its single-stream decode is faster than a dense model of similar size.
On the community SparkBench leaderboard (PBM methodology, 4k context, single GB10), this model with NVFP4 + MTP measures 86.3 tokens/s single-stream. We haven't published our own tuned numbers yet.
A dedicated NVIDIA DGX Spark (GB10, 128 GB unified memory) rented by the hour, week, month or year, running Qwen3.6 35B A3B behind an OpenAI-compatible endpoint on your own machine, hosted in the EU with zero prompt retention.
Hardware from €0.85/hour, by the hour, week, month or year; the managed service is quoted per deployment. Request deployment and we confirm both in writing.
Qwen3.8 27B is available now on the serverless API at €0.32 per 1M input tokens and €2.00 per 1M output tokens (launch pricing).
Explore