Model catalog
Know what you can build against before you write the first request.
Kimi K3 is Arnict's launch model. Its identifier, parameter scale, hosting location, token rates, and availability are published here; future models remain Upcoming until explicitly released.
Kimi K3
kimi-k3
Total params
2.8T
Active params
104B
Catalog position
First model
Inference
Coming soon
Kimi K3 is the sole designated launch model and the first model in the Arnict catalog. It is directly hosted on Arnict-operated US-based servers — the model-serving path is owned end to end, with no third-party inference host in between.
Use the API identifier kimi-k3 in chat completion requests. Inference availability is reported on the status page.
Token rates · USD per 1M
| Input tokens | $2.50 |
| Cached input tokens | $0.25 |
| Output tokens | $12.50 |
Hosted inference path
- → Arnict-operated US-based servers
- → OpenAI-compatible request shape
- → Metered at published rates
- → Not routed through third parties
Roadmap
Future models, in priority order.
Every item below is Upcoming until explicitly released. No availability date is implied by catalog position.
| Priority | Model | Scale | Status |
|---|---|---|---|
| 01 | Qwen3.8 MAX | 2.4T total parameters | Upcoming |
| 02 | DeepSeek V4 Pro | 1.6T total parameters · 49B active parameters | Upcoming |
| 03 | DeepSeek V4 Flash | 284B total parameters · 13B active parameters | Upcoming |
| 04 | Step 3.7 Flash | 196B total parameters | Upcoming |
| 05 | Affordable model range | Additional smaller parameter sizes | Upcoming |
EU-hosted models
An upcoming infrastructure option for selected models. Not available at launch, and no additional hosting location is claimed beyond US-based Arnict-operated servers.