Open-weight models
Coming

Open weights, the same endpoint.

Llama, Qwen, DeepSeek and Mixtral are staged in the catalogue, ready to serve through the same OpenAI-compatible endpoint, the same routing rules and the same per-customer billing as everything else. They are not live yet, and there is no date.

There are no prices on this page because none have been set. On odnoga a model without a price cannot be called — that rule is what stops a number appearing here before it is real.

What is staged

Llama 3.3 70B

Soon
Context
131K
Licence
Llama

Staged with: Cerebras, DeepInfra, Fireworks AI

Llama 3.3 70B Turbo

Soon
Context
131K
Licence
Llama

Staged with: Together AI

Llama 4 Maverick 17B

Soon
Context
131K
Licence
Llama

Staged with: Cerebras

Llama 4 Scout 17B

Soon
Context
131K
Licence
Llama

Staged with: Cerebras

Qwen 2.5 72B

Soon
Context
131K
Licence
Apache-2.0

Staged with: DeepInfra, Fireworks AI

Qwen 2.5 72B Turbo

Soon
Context
131K
Licence
Apache-2.0

Staged with: Together AI

Qwen 3 32B

Soon
Context
131K
Licence
Apache-2.0

Staged with: Cerebras

DeepSeek V3

Soon
Context
66K
Licence
DeepSeek

Staged with: DeepInfra, Fireworks AI, Together AI

Mixtral 8x22B

Soon
Context
66K
Licence
Apache-2.0

Staged with: Fireworks AI, Together AI

Mixtral 8x7B

Soon
Context
33K
Licence
Apache-2.0

Staged with: DeepInfra

Several providers can serve the same open weights, which is the point: routing picks one, and a fallback can move to another without your code changing. The context window shown is the largest any of them offers.

What you can use today

The catalogue is live, priced and callable now — every model in it has a published price, because that is the condition for serving it at all.

Build on the endpoint now, switch weights later.

Models change by routing rule, not by redeploy — so anything that lands later is a setting, not a migration.