Groq

Fast open-weight LLMs (OpenAI gpt-oss on Groq LPUs) for text work: summarize, extract JSON, classify, translate.

Provider id: groq
Tools: 4, at $0.005 per call
Categories: Text AI
Homepage: groq.com
Provider docs: console.groq.com/docs

Tool	Price	What it does
`groq/classify`	$0.005	Put text into exactly one of your labels, with a confidence and a one-sentence reason.
`groq/extract`	$0.005	Pull structured fields out of text as a JSON object: list the keys and types, or describe what you want.
`groq/summarize`	$0.005	Summarise up to 20k characters of text into a short paragraph plus key points, in about a second.
`groq/translate`	$0.005	Translate up to 5,000 characters into any language, keeping formatting, and name the source language.

Inputs are JSON bodies, and unknown fields are rejected with 422 invalid_input (not charged). POST /v1/inspect with a tool's id returns the same input schema, plus the output schema.

Groq Classify

groq/classify · $0.005 per call (standard) · Text AI

Put text into exactly one of your labels, with a confidence and a one-sentence reason.

Zero-shot classification: give the text and 2–25 labels (sentiment, topic, intent, priority, language register...) and get back one label from your list (schema-enforced), a 0–1 confidence and the reason. Add criteria to define borderline cases. One label per call; for several attributes at once use groq/extract with boolean or string fields.

Input

Field	Type	Required	Description
`text`	string	yes	The text to label. 1–10000 characters.
`labels`	array of string	yes	The allowed labels; exactly one is chosen. 2–25 items.
`criteria`	string	no	How to decide, e.g. 'urgent = needs a reply today'. At most 500 characters.

Example

Input

{
  "text": "My card was charged twice for the same order and I need the duplicate refunded today.",
  "labels": [
    "billing",
    "shipping",
    "technical issue",
    "account access",
    "other"
  ]
}

Run it with the CLI

node akashi.mjs run groq/classify --input '{"text":"My card was charged twice for the same order and I need the duplicate refunded today.","labels":["billing","shipping","technical issue","account access","other"]}'

Groq Extract JSON

groq/extract · $0.005 per call (standard) · Text AI, Content Extraction

Pull structured fields out of text as a JSON object: list the keys and types, or describe what you want.

Reads text (an email, invoice, bio, product page, transcript) and returns a JSON object. With fields, the answer has exactly those keys with those types (schema-enforced) and null for anything the text does not say; with only an instruction, the model chooses the shape. It never invents values that are not in the text. It does not fetch URLs: read the page first with jina/read.

Input

Field	Type	Required	Description
`text`	string	yes	The text to read. 1–20000 characters.
`fields`	array of object (FieldSpec)	no	The keys you want back, with types. Missing values come back as null. At most 30 items.
`instruction`	string	no	Plain-language request instead of (or as well as) fields, e.g. 'every person mentioned, with their role'. At most 1000 characters.

Each FieldSpec object:

Field	Type	Required	Description
`name`	string	yes	JSON key, e.g. 'invoice_total'. Pattern `^[A-Za-z_][A-Za-z0-9_]{0,63}$`.
`type`	string	no	Value type; string_list for a list of strings. One of `string`, `number`, `integer`, `boolean`, `string_list`. Default `"string"`.
`description`	string	no	What to put here, e.g. 'total including tax, in USD'. At most 200 characters.

Example

Input

{
  "text": "Hi team, invoice INV-2291 from Northwind Traders is due on 14 November 2026. The total is $1,284.50 including tax. Contact: Maria Lopez, maria@northwind.example.",
  "fields": [
    {
      "name": "invoice_number",
      "type": "string"
    },
    {
      "name": "vendor",
      "type": "string"
    },
    {
      "name": "due_date",
      "type": "string",
      "description": "ISO date"
    },
    {
      "name": "total_usd",
      "type": "number"
    },
    {
      "name": "contact_email",
      "type": "string"
    }
  ]
}

Run it with the CLI

node akashi.mjs run groq/extract --input '{"text":"Hi team, invoice INV-2291 from Northwind Traders is due on 14 November 2026. The total is $1,284.50 including tax. Contact: Maria Lopez, maria@northwind.example.","fields":[{"name":"invoice_number","type":"string"},{"name":"vendor","type":"string"},{"name":"due_date","type":"string","description":"ISO date"},{"name":"total_usd","type":"number"},{"name":"contact_email","type":"string"}]}'

Groq Summarize

groq/summarize · $0.005 per call (standard) · Text AI

Summarise up to 20k characters of text into a short paragraph plus key points, in about a second.

Condenses text you already have into a paragraph (≤ 120 words) and up to 10 one-line key points, using only what the text says; set focus to steer it. It does not fetch URLs: read a page first with jina/read. It does not answer questions from the web: use akashi/answer. Text over 20,000 characters is rejected, so split long documents.

Input

Field	Type	Required	Description
`text`	string	yes	The text to summarise (an article, transcript, report, thread). 40–20000 characters.
`max_points`	integer	no	How many key points at most. From 1 to 10. Default `5`.
`focus`	string	no	What the reader cares about, e.g. 'pricing changes' or 'risks'. At most 300 characters.

Example

Input

{
  "text": "The transistor was invented at Bell Labs in 1947 by John Bardeen and Walter Brattain, working under William Shockley. Their point-contact device used a germanium crystal to amplify a signal, replacing bulky and power-hungry vacuum tubes. Shockley followed with the more robust junction transistor in 1948. The three shared the 1956 Nobel Prize in Physics. Transistors made portable radios, computers and eventually integrated circuits possible; a modern phone chip holds billions.",
  "max_points": 3
}

Run it with the CLI

node akashi.mjs run groq/summarize --input '{"text":"The transistor was invented at Bell Labs in 1947 by John Bardeen and Walter Brattain, working under William Shockley. Their point-contact device used a germanium crystal to amplify a signal, replacing bulky and power-hungry vacuum tubes. Shockley followed with the more robust junction transistor in 1948. The three shared the 1956 Nobel Prize in Physics. Transistors made portable radios, computers and eventually integrated circuits possible; a modern phone chip holds billions.","max_points":3}'

Groq Translate

groq/translate · $0.005 per call (standard) · Text AI, Words & Language

Translate up to 5,000 characters into any language, keeping formatting, and name the source language.

Translates text with an LLM, preserving markdown, line breaks, numbers, names and URLs, and reports the detected source language. Good for most major languages; quality drops for rare ones, so do not use it for legal or medical documents without a human check. Over 5,000 characters is rejected: split the text.

Related: groq/summarize.

Input

Field	Type	Required	Description
`text`	string	yes	The text to translate. 1–5000 characters.
`target_language`	string	yes	Language to translate into, by name or code: 'French', 'pt-BR', 'Yoruba'. 2–40 characters.
`source_language`	string	no	The text's language if known; detected otherwise. 2–40 characters.

Example

Input

{
  "text": "The meeting moved to Thursday at 3 pm. Please bring the signed contract.",
  "target_language": "French"
}

Run it with the CLI

node akashi.mjs run groq/translate --input '{"text":"The meeting moved to Thursday at 3 pm. Please bring the signed contract.","target_language":"French"}'

On this page