Model API · Available
When this offering is available: Qwen3.8 Flash Next, the Flash size of the Qwen3.8 generation, through an OpenAI-compatible endpoint on hardware AxForge owns in Stockholm. It reads up to 500,000 tokens at once, takes images and calls your tools, and your prompts stay in the EU with zero prompt retention.
Why AxForge
| OpenAI-compatible | Chat completions, streaming, tool calls and usage reporting follow the OpenAI API shape. Point your SDK at api.axforge.ai/v1 and pick the model qwen3.8-flash-next. |
|---|---|
| Long context, images too | It reads up to 500,000 tokens at once, whole codebases or long documents, and takes images in the same request. |
| Zero prompt retention | Prompts and completions are processed in memory in Stockholm, Sweden. Not logged, not retained, never used for training. |
Specifications
| Model | Qwen3.8 Flash Next: mixture of experts, 125B parameters with 6B active; Qwen Community 1.0 and NVIDIA Open Model licences |
|---|---|
| API model name | qwen3.8-flash-next |
| Context window | 500,000 tokens |
| Quantization | NVFP4 |
| Capabilities | Chat completions · streaming · tool calls · image input · thinking when you ask |
| Hardware | One NVIDIA DGX Spark (GB10), owned and operated by AxForge |
| Region | Stockholm, Sweden (eu-se-1) |
| Input price | €0.12 / 1M tokens |
| Cached input price | €0.06 / 1M tokens |
| Output price | €0.62 / 1M tokens |
List prices in EUR.
How it works
| 1 | Confirm this offering is available in the console before following the API setup steps below; otherwise add a public request. A request does not activate an endpoint. |
|---|---|
| 2 | Sign in to AxForge. Your API key is created in the console. |
| 3 | Set base_url to https://api.axforge.ai/v1 in the OpenAI SDK you already use. |
| 4 | Send your first request with model qwen3.8-flash-next. |
| 5 | Watch usage in the console; every new account starts with free tokens. |
Try Qwen3.8 Flash Next or sign in to start.
FAQ
Yes. Chat completions, streaming, tool calls and usage reporting follow the OpenAI API shape, so the official SDKs work by changing base_url.
When you ask it to. Send enable_thinking, or a reasoning_effort of low, medium or xhigh (high and max are read as xhigh). Without them it answers directly.
Yes. Send an image in the message as with other vision models; see Sending images in the docs.
Up to 500,000 tokens in one request: a long document, a codebase or many files together.
When available, check current pricing in the console. €0.12 per million input tokens, €0.06 per million cached input tokens and €0.62 per million output tokens. No subscription, no minimum.
In Stockholm, Sweden (eu-se-1), on an NVIDIA DGX Spark AxForge owns and operates. TLS terminates in the EU.
No. Prompts and completions are processed in memory and never persisted or used for training.