Models

Novin Cloud provides access to dozens of language models from various providers. All of them are available through a single API with a single key.

The live model list is available in the AI → Models section of the user console.

Viewing models in the console

To see the model list in the console, you must first have an active API key and select it from the menu at the top of the page. The model list is shown based on that key's access.

Model ID

Every model has a Model ID used in requests. IDs follow the pattern provider/model-name:

{
  "model": "anthropic/claude-sonnet-4-5",
  "messages": [...]
}

warning

The model ID must match the list exactly. gpt-4o without the openai/ prefix is not valid and returns a 404 error.

To get the exact list of IDs programmatically:

curl https://iapi.novin.cloud/v1/models \
  -H "Authorization: Bearer $NOVIN_API_KEY"

Model specifications

Each model's card in the console shows the following information:

Context window

The maximum number of tokens the model can process in a single request — including the total of the input message, conversation history, and generated response.

If the total exceeds this limit, the request fails with an error. For long conversations or processing large documents, choose a model with a larger context window.

Window size	Approximate use case
64K tokens	Everyday conversations, short text
128K tokens	Long documents, large codebases
200K tokens	Multiple documents at once, full projects
1M tokens	Books, entire code repositories

What is a token?

A token is the unit of text processing used by models. Roughly, one token is equivalent to three to four characters of English text. Persian text usually consumes more tokens than an equivalent amount of English text.

Capabilities

Each model may support the following capabilities:

Capability	Description
Vision	The ability to receive and analyze images alongside text
Function Calling	The ability to call tools and functions you define — the foundation for building agents
Streaming	Sending the response incrementally, word by word

Input and output pricing

Each model has two separate prices:

• Input price — for the tokens you send to the model (message and history)
• Output price — for the tokens the model generates

The output price is usually three to five times the input price.

Up-to-date pricing

Prices may change. Always check the current price for each model on the Models page in the console. The figures on this page are provided only for relative comparison between models.

Providers and model families

Anthropic — the Claude family

Claude models excel at reasoning, long-form writing, and coding, and have a 200K-token context window.

Model	Notes
anthropic/claude-opus-4	The most capable model in the family, for complex tasks
anthropic/claude-sonnet-4-5	A good balance of quality and cost
anthropic/claude-haiku-4-5	Fast and inexpensive, for simple, high-volume tasks
anthropic/claude-3.5-haiku	The previous generation of the fast model

OpenAI — the GPT and o-series families

Model	Notes
openai/gpt-4o	A multimodal model with image support
openai/gpt-4o-mini	A lightweight, inexpensive version of gpt-4o
openai/gpt-4-turbo	The previous generation, still capable
openai/o1, openai/o3	Reasoning models for complex math and logic problems
openai/o4-mini	A lighter reasoning model

Reasoning models

The o-series models work through reasoning steps before responding. They're more accurate for complex math, logic, and programming problems, but slower and more expensive. Don't use them for simple questions.

Google — the Gemini family

The defining feature of this family is its very large context window (over one million tokens) — suited to processing large documents and code repositories.

Model	Notes
google/gemini-2.5-pro	The most capable model in the family
google/gemini-2.5-flash	Fast, with the same large context window
google/gemini-2.5-flash-lite	The lightest and cheapest option

Meta — the Llama family

Open-source models with good quality and reasonable pricing.

Model	Notes
meta-llama/llama-3.3-70b-instruct	Good quality at a balanced cost
meta-llama/llama-3.1-405b-instruct	The largest model in the family
meta-llama/llama-3.2-11b-vision-instruct	Supports images

DeepSeek

Cost-effective models focused on reasoning and coding.

Model	Notes
deepseek/deepseek-chat-v3.1	The cheapest option for general tasks
deepseek/deepseek-r1	A reasoning model
deepseek/deepseek-r1-distill-llama-70b	A distilled, lighter version of the reasoning model

Mistral

Model	Notes
mistralai/mistral-large-2411	The flagship model
mistralai/codestral-2508	Specialized for coding, 256K-token context window
mistralai/mistral-nemo	Lightweight and inexpensive

Qwen

Model	Notes
qwen/qwen3-235b-a22b	The largest model in the family
qwen/qwen3-coder	Specialized for coding
qwen/qwen-plus	Cost-effective for everyday tasks

Model selection guide

Your need	Recommendation
Simple, high-volume tasks (classification, short summarization)	Lightweight models: gpt-4o-mini, claude-haiku-4-5, gemini-2.5-flash-lite
General conversation with good quality	gpt-4o, claude-sonnet-4-5, gemini-2.5-flash
Long-form, high-quality writing	claude-sonnet-4-5, claude-opus-4
Coding	codestral-2508, qwen3-coder, claude-sonnet-4-5
Complex math and logic problems	openai/o1, openai/o3, deepseek-r1
Processing very large documents	The gemini-2.5 family (million-token window)
Image analysis	Models with vision capability
Lowest cost	deepseek-chat-v3.1, mistral-nemo, qwen-plus

A practical approach to choosing

Start with a mid-tier model. If the response quality is sufficient, try a cheaper model to find the lowest acceptable cost. If it's not sufficient, move up to a more capable model. Testing with the console chat is the fastest way to compare.

Filtering and search

On the models page in the console, you can filter the list by provider and capability, or search by model name.

Next steps

• Chat — quickly try out models without writing code
• API Reference — using models in code
• Usage & Billing — comparing the actual cost of models