Ask AI

Perplexity

Perplexity models combine language generation with live web search. Use them when you need sourced answers, current information, citation metadata, or research-style synthesis through AvalAI's OpenAI-compatible API.

AvalAI currently exposes the supported Perplexity LLMs through /v1/chat/completions and exposes direct search through v1/search. Responses examples are included as migration patterns for routes/models that explicitly support /v1/responses; keep native Sonar calls on Chat Completions when you need Perplexity-specific fields such as citations, search_results, or search_context_size.

Available Models

sonar - fast answers with reliable search results
sonar-pro - advanced search with richer results
sonar-reasoning - quick chain-of-thought-style reasoning with live search
sonar-reasoning-pro - deeper reasoning with broader search
sonar-deep-research - comprehensive research across many sources
perplexity-search - direct search results without an LLM response

API Endpoint Support

Capability	Recommended endpoint	Notes
Native Perplexity LLM answers	`/v1/chat/completions`	Best for Sonar models, citations, search metadata, and Perplexity-specific response fields.
Responses-style text workflows	`/v1/responses`	Use only with a model/route that supports Responses; map `messages` to `input` and read `output_text`.
Direct search only	`/v1/search`	Use `perplexity-search` when you need search results without generated prose.

sonar

Fast answers with reliable search results. Use sonar for quick grounded answers, news checks, definitions, and lightweight browsing tasks.

Features

Non-reasoning model optimized for speed and cost
128K context window
Live web search with citations and search-result metadata
Good default for short factual questions and summaries

Pricing

Type	Cost
Input tokens	$1.00 / 1M tokens
Cached input tokens	$0.50 / 1M tokens
Output tokens	$1.00 / 1M tokens
Search context: low	$5 / 1K requests
Search context: medium	$8 / 1K requests
Search context: high	$12 / 1K requests

Example

BashPythonJavaScript

bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "sonar",
    "messages": [
      {
        "role": "user",
        "content": "What is the latest news in AI research?"
      }
    ]
  }'

python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir/v1",
)

response = client.chat.completions.create(
    model="sonar",
    messages=[{"role": "user", "content": "What is the latest news in AI research?"}],
)

print(response.choices[0].message.content)
print(getattr(response, "citations", []))

javascript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

const response = await client.chat.completions.create({
  model: "sonar",
  messages: [{ role: "user", content: "What is the latest news in AI research?" }],
});

console.log(response.choices[0].message.content);
console.log(response.citations ?? []);

Responses API migration pattern

Use this shape only when the selected AvalAI model supports /v1/responses. It does not expose Perplexity-specific citation metadata; keep the Sonar Chat Completions request when you need native search fields.

PythonJavaScriptBash

python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir/v1",
)

response = client.responses.create(
    model=os.getenv("AVALAI_RESPONSES_MODEL", "gpt-5.5"),
    instructions="Answer with concise citations when available.",
    input="What is the latest news in AI research?",
)

print(response.output_text)

javascript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

const response = await client.responses.create({
  model: process.env.AVALAI_RESPONSES_MODEL ?? "gpt-5.5",
  instructions: "Answer with concise citations when available.",
  input: "What is the latest news in AI research?",
});

console.log(response.output_text);

bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "instructions": "Answer with concise citations when available.",
    "input": "What is the latest news in AI research?"
  }'

messages → input
system or policy prompt → instructions
choices[0].message.content → response.output_text
Perplexity citations / search_results → keep Chat Completions or call /v1/search

sonar-pro

Advanced search with enhanced result quality. Use sonar-pro for research questions that need broader retrieval, comparison across sources, or longer synthesis.

Features

200K input context
More search results than standard Sonar
Better fit for market analysis, research summaries, and multi-source synthesis
Includes live search metadata and citations

Pricing

Type	Cost
Input tokens	$3.00 / 1M tokens
Cached input tokens	$1.50 / 1M tokens
Output tokens	$15.00 / 1M tokens
Search context: low	$6 / 1K requests
Search context: medium	$10 / 1K requests
Search context: high	$14 / 1K requests

Example

BashPythonJavaScript

bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "sonar-pro",
    "messages": [
      {
        "role": "user",
        "content": "Analyze the competitive landscape for AI search engines in 2026."
      }
    ]
  }'

python

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AVALAI_API_KEY"],
    base_url="https://api.avalai.ir/v1",
)

response = client.chat.completions.create(
    model="sonar-pro",
    messages=[
        {
            "role": "user",
            "content": "Analyze the competitive landscape for AI search engines in 2026.",
        }
    ],
)

print(response.choices[0].message.content)

javascript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.AVALAI_API_KEY,
  baseURL: "https://api.avalai.ir/v1",
});

const response = await client.chat.completions.create({
  model: "sonar-pro",
  messages: [
    { role: "user", content: "Analyze the competitive landscape for AI search engines in 2026." },
  ],
});

console.log(response.choices[0].message.content);

Responses API migration pattern

PythonJavaScriptBash

python

response = client.responses.create(
    model=os.getenv("AVALAI_RESPONSES_MODEL", "gpt-5.5"),
    instructions="Create a sourced competitive analysis.",
    input="Analyze the competitive landscape for AI search engines in 2026.",
)
print(response.output_text)

javascript

const response = await client.responses.create({
  model: process.env.AVALAI_RESPONSES_MODEL ?? "gpt-5.5",
  instructions: "Create a sourced competitive analysis.",
  input: "Analyze the competitive landscape for AI search engines in 2026.",
});
console.log(response.output_text);

bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "instructions": "Create a sourced competitive analysis.",
    "input": "Analyze the competitive landscape for AI search engines in 2026."
  }'

Use Chat Completions with sonar-pro when you need native Perplexity citations and search-result fields.

sonar-reasoning

Fast reasoning with real-time search. Use sonar-reasoning when the answer benefits from visible stepwise synthesis but still needs low-latency search.

Features

Live web search with reasoning-oriented answer synthesis
128K context window
Good for quick fact-checking, current-event explanations, and multi-step Q&A
Lower output cost than the Pro reasoning model

Pricing

Type	Cost
Input tokens	$1.00 / 1M tokens
Cached input tokens	$0.50 / 1M tokens
Output tokens	$5.00 / 1M tokens
Search context: low	$5 / 1K requests
Search context: medium	$8 / 1K requests
Search context: high	$14 / 1K requests

Example

BashPythonJavaScript

bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "sonar-reasoning",
    "messages": [
      {
        "role": "user",
        "content": "Explain the main tradeoffs of open-weight AI models for startups."
      }
    ]
  }'

python

response = client.chat.completions.create(
    model="sonar-reasoning",
    messages=[
        {
            "role": "user",
            "content": "Explain the main tradeoffs of open-weight AI models for startups.",
        }
    ],
)
print(response.choices[0].message.content)

javascript

const response = await client.chat.completions.create({
  model: "sonar-reasoning",
  messages: [
    { role: "user", content: "Explain the main tradeoffs of open-weight AI models for startups." },
  ],
});
console.log(response.choices[0].message.content);

Responses API migration pattern

PythonJavaScriptBash

python

response = client.responses.create(
    model=os.getenv("AVALAI_RESPONSES_MODEL", "gpt-5.5"),
    reasoning={"effort": "medium"},
    input="Explain the main tradeoffs of open-weight AI models for startups.",
)
print(response.output_text)

javascript

const response = await client.responses.create({
  model: process.env.AVALAI_RESPONSES_MODEL ?? "gpt-5.5",
  reasoning: { effort: "medium" },
  input: "Explain the main tradeoffs of open-weight AI models for startups.",
});
console.log(response.output_text);

bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "reasoning": {"effort": "medium"},
    "input": "Explain the main tradeoffs of open-weight AI models for startups."
  }'

Map Chat reasoning examples to the Responses reasoning.effort object when the selected model supports it.

sonar-reasoning-pro

Advanced reasoning with comprehensive search. Use sonar-reasoning-pro for research briefs, high-stakes comparisons, and multi-source answers where depth matters more than latency.

Features

Reasoning-oriented synthesis with live search
128K context window
Better fit for complex research and strategy questions
Higher search-context pricing for richer retrieval

Pricing

Type	Cost
Input tokens	$2.00 / 1M tokens
Cached input tokens	$1.00 / 1M tokens
Output tokens	$8.00 / 1M tokens
Search context: low	$6 / 1K requests
Search context: medium	$10 / 1K requests
Search context: high	$14 / 1K requests

Example

BashPythonJavaScript

bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "sonar-reasoning-pro",
    "messages": [
      {
        "role": "user",
        "content": "Compare regulatory approaches to AI safety in the US, EU, and UK."
      }
    ]
  }'

python

response = client.chat.completions.create(
    model="sonar-reasoning-pro",
    messages=[
        {
            "role": "user",
            "content": "Compare regulatory approaches to AI safety in the US, EU, and UK.",
        }
    ],
)
print(response.choices[0].message.content)

javascript

const response = await client.chat.completions.create({
  model: "sonar-reasoning-pro",
  messages: [
    { role: "user", content: "Compare regulatory approaches to AI safety in the US, EU, and UK." },
  ],
});
console.log(response.choices[0].message.content);

Responses API migration pattern

PythonJavaScriptBash

python

response = client.responses.create(
    model=os.getenv("AVALAI_RESPONSES_MODEL", "gpt-5.5"),
    reasoning={"effort": "high"},
    input="Compare regulatory approaches to AI safety in the US, EU, and UK.",
)
print(response.output_text)

javascript

const response = await client.responses.create({
  model: process.env.AVALAI_RESPONSES_MODEL ?? "gpt-5.5",
  reasoning: { effort: "high" },
  input: "Compare regulatory approaches to AI safety in the US, EU, and UK.",
});
console.log(response.output_text);

bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "reasoning": {"effort": "high"},
    "input": "Compare regulatory approaches to AI safety in the US, EU, and UK."
  }'

Use reasoning.effort to tune quality, latency, and cost on supported Responses models.

sonar-deep-research

Comprehensive research across many sources. Use sonar-deep-research for long-form reports, source gathering, and questions where citation breadth is the product value.

Features

Deep research workflow over live web sources
128K context window
Citation-aware output with additional pricing dimensions for reasoning and citations
Best for reports, market research, policy reviews, and due-diligence drafts

Pricing

Type	Cost
Input tokens	$2.00 / 1M tokens
Cached input tokens	$1.00 / 1M tokens
Output tokens	$8.00 / 1M tokens
Reasoning output	$3.00 / 1M tokens
Citations	$2.00 / 1M citations
Search context	$5 / 1K requests

Example

BashPythonJavaScript

bash

curl https://api.avalai.ir/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "sonar-deep-research",
    "messages": [
      {
        "role": "user",
        "content": "Create a sourced brief on AI adoption in healthcare systems."
      }
    ]
  }'

python

response = client.chat.completions.create(
    model="sonar-deep-research",
    messages=[
        {
            "role": "user",
            "content": "Create a sourced brief on AI adoption in healthcare systems.",
        }
    ],
)
print(response.choices[0].message.content)
print(getattr(response, "citations", []))

javascript

const response = await client.chat.completions.create({
  model: "sonar-deep-research",
  messages: [
    { role: "user", content: "Create a sourced brief on AI adoption in healthcare systems." },
  ],
});
console.log(response.choices[0].message.content);
console.log(response.citations ?? []);

Responses API migration pattern

PythonJavaScriptBash

python

response = client.responses.create(
    model=os.getenv("AVALAI_RESPONSES_MODEL", "gpt-5.5"),
    reasoning={"effort": "high"},
    instructions="Write a concise, source-aware research brief.",
    input="Create a sourced brief on AI adoption in healthcare systems.",
    store=False,
)
print(response.output_text)

javascript

const response = await client.responses.create({
  model: process.env.AVALAI_RESPONSES_MODEL ?? "gpt-5.5",
  reasoning: { effort: "high" },
  instructions: "Write a concise, source-aware research brief.",
  input: "Create a sourced brief on AI adoption in healthcare systems.",
  store: false,
});
console.log(response.output_text);

bash

curl https://api.avalai.ir/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "model": "gpt-5.5",
    "reasoning": {"effort": "high"},
    "instructions": "Write a concise, source-aware research brief.",
    "input": "Create a sourced brief on AI adoption in healthcare systems.",
    "store": false
  }'

For true Perplexity deep-research citations, keep the native Chat Completions request. Use Responses when your workflow needs Responses state, tools, or output_text handling on a supported model.

Direct Search API Without LLM

Use perplexity-search with v1/search when you want search results and snippets without model-generated prose.

Request Option 1: Tool In URL

bash

curl "https://api.avalai.ir/v1/search/perplexity-search?q=latest%20AI%20research" \
  -H "Authorization: Bearer $AVALAI_API_KEY"

Request Option 2: Tool In Body

bash

curl https://api.avalai.ir/v1/search \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $AVALAI_API_KEY" \
  -d '{
    "tool": "perplexity-search",
    "query": "latest AI research"
  }'

Parameters

Parameter	Type	Required	Description
`tool`	string	Yes	Use `perplexity-search` when calling `/v1/search` with a JSON body.
`query` or `q`	string	Yes	Search query text.
`search_context_size`	string	No	Optional search depth such as `low`, `medium`, or `high` when supported.

Implementation Notes

Log citations, search_results, usage, and x-request-id for debugging and cost review.
Treat live-search answers as time-sensitive; re-run important queries before showing stale information.
For strict privacy workflows, review the data sent to search providers and avoid including secrets, credentials, or unrelated PII.
For Responses migrations, confirm the selected model supports /v1/responses; not every provider/model exposes the same feature set.

Perplexity ​

Available Models ​

API Endpoint Support ​

sonar ​

Features ​

Pricing ​

Example ​

sonar-pro ​

Features ​

Pricing ​

Example ​

sonar-reasoning ​

Features ​

Pricing ​

Example ​

sonar-reasoning-pro ​

Features ​

Pricing ​

Example ​

sonar-deep-research ​

Features ​

Pricing ​

Example ​

Direct Search API Without LLM ​

Request Option 1: Tool In URL ​

Request Option 2: Tool In Body ​

Parameters ​

Implementation Notes ​

Related Resources ​

Perplexity

Available Models

API Endpoint Support

sonar

Features

Pricing

Example

sonar-pro

Features

Pricing

Example

sonar-reasoning

Features

Pricing

Example

sonar-reasoning-pro

Features

Pricing

Example

sonar-deep-research

Features

Pricing

Example

Direct Search API Without LLM

Request Option 1: Tool In URL

Request Option 2: Tool In Body

Parameters

Implementation Notes

Related Resources