Models and prices¶
The catalog is evaluated for the authenticated key. A model shown in another profile or in an example is not a promise that it is enabled for your key.
Use the bot's Models screen for a human-readable list and this request for the machine-readable source:
curl https://api.nexus-hub.tech/v1/models \
-H "Authorization: Bearer $NEXUS_API_KEY"
The response lists the model ids this key can call. It does not list route
families. owned_by is a coarse label, not a protocol selector: copy id
exactly and choose the protocol in the client. GET /v1/models/{model} checks
one id with the same visibility rules.
Route families and verified parameters are on the public catalog:
curl https://api.nexus-hub.tech/v1/catalog
Read endpoint_families for the id you copied. A model can be allowed for chat
and not for responses. If the client then returns unsupported_model, switch
the protocol or pick another id from GET /v1/models.
Prices are estimates and can change. The bot displays them in the selected currency; the API balance is maintained in RUB. See Pricing and balance for usage and top-up behaviour.
Families¶
- OpenAI-compatible text and response models
- Anthropic-compatible message models
- Gemini-compatible generation and embedding models
- Additional families shown by the current key-specific catalog
Choose a model only after reading the current response. Do not hard-code a model that is not returned for the key.
Context windows¶
Every model currently returned by GET /v1/models for a Nexus key has a declared context window of 1,050,000 tokens. Use that number when a client asks for a context size. A model that is absent from the key's list is not available on that key.
The price multipliers in the next section change the price of a large GPT request. They do not shrink the context window.
Dynamic pricing¶
For certain models, request pricing depends on the input context size or the request admission time. Multipliers apply to the active base model price displayed in Nexus.
GPT models¶
For GPT models (with ID gpt-*), pricing depends on the total input context size. The context size includes all input tokens, including tokens read from cache (input_tokens + cached_input_tokens).
| Context | Input | Output | Cache Read | Cache Write |
|---|---|---|---|---|
| ≤ 272,000 tokens | ×1 | ×1 | ×1 | ×1 |
| > 272,000 tokens | ×2 | ×1.5 | ×2 | ×2 |
DeepSeek models¶
For DeepSeek models (with ID deepseek-*), a time-based multiplier applies in the UTC timezone (corresponding to peak windows 09:00–12:00 and 14:00–18:00 Asia/Shanghai). The window is determined by the request start (admission) time.
| Period (UTC) | Multiplier | Window in Asia/Shanghai |
|---|---|---|
| Monday–Friday 01:00–04:00 UTC | ×2 | 09:00–12:00 Asia/Shanghai |
| Monday–Friday 06:00–10:00 UTC | ×2 | 14:00–18:00 Asia/Shanghai |
| All other times and weekends | ×1 | Standard rates |
See the parameter matrix for generation capabilities.