ASasknews
chat

Deep research into real-time news, archive news, and Google.

Changed on
post/v1/chat/deepnews

Request

Request body

model'o1-mini' | 'o1' | 'o3-mini' | 'o3' | 'gpt-4o-mini' | 'gpt-4o' | 'gpt-4.1-2025-04-14' | 'gpt-5-mini' | 'gpt-5' | 'command-nightly' | 'claude-3-5-sonnet-20240620' | 'claude-3-5-sonnet-latest' | 'claude-3-7-sonnet-latest' | 'claude-3-7-sonnet-20250219' | 'claude-3-7-sonnet-nothinking' | 'claude-sonnet-4-20250514' | 'claude-opus-4-20250514' | 'claude-sonnet-4-5-20250929' | 'claude-opus-4-5-20251101' | 'claude-sonnet-4-6' | 'claude-sonnet-5' | 'claude-sonnet-5-5' | 'claude-opus-4-8' | 'claude-opus-5' | 'claude-opus-5-5' | 'claude-opus-4-6' | 'claude-fable-5' | 'claude-fable-5-1' | 'gpt-5.6-terra' | 'gpt-5.6-sol' | 'gpt-6-sol' | 'gpt-6.1-sol' | 'gpt-6-astra' | 'deepseek' | 'deepseek-basic' | 'deepseek-r1-0528' | 'meta-llama/Meta-Llama-3-8B-Instruct' | 'meta-llama/Meta-Llama-3.1-405B-Instruct' | 'meta-llama/Meta-Llama-3.3-70B-Instruct' | 'meta-llama/Llama-4-Maverick-17B-128E-Instruct' | 'gemini-2.5-flash' | 'gemini-2.5-pro' | 'gemini-2.5-pro-preview-03-25' | 'gemini-3-pro' | 'gemini-3.1-pro' | 'gemini-4-argon' | 'gemini-3-flash' | 'gemini-3.5-flash' | 'gemini-3.6-flash' | 'open-source-best' | 'kimi-k2p5' | 'kimi-k3'
filter_paramsobject nullable

Filter parameters to apply to the AskNews search within DeepNews.

search_depthinteger

The search depth for deep research. Higher values mean more thorough research.

max_depthinteger

The maximum research depth allowed.

inline_citations'markdown_link' | 'numbered' | 'none'

How to format inline citations in the response. Defaults to: markdown_link. Available options: markdown_link, numbered, none

start_citation_numberinteger

Starting number for inline citations. Offsets fetched source citation keys. Useful if you are providing the agent outside sources with numbered citation keys.

return_sourcesboolean

Whether return a list of sources used by the agent.

only_cited_sourcesboolean

Whether to return only of sources that are cited/referenced in the generated response content.

append_referencesboolean

Whether to append a 'Cited Articles' section at the end of the response.

journalist_modeboolean

Whether to use journalist mode for more factual reporting.

conversational_awarenessboolean

Whether to use conversational awareness.

include_entitiesboolean

Whether to provide extracted entities to the agent.

include_graphsboolean

Whether to provide knowledge graphs to the agent

include_coordinatesboolean

Whether to provide geo coordinates to the agent.

asknews_watermarkboolean

Append 'Generated by AskNews AI' watermark

streamboolean

Whether to stream the response as server-sent events.

temperaturenumber

The temperature of the DeepNews agent model. Controls randomness in the output. Higher values (e.g., 1.0) make output more random, lower values (e.g., 0.1) make it more deterministic.

top_pnumber

Nucleus sampling parameter. Only tokens with cumulative probability up to top_p are considered.

ninteger

Number of completions to generate.

max_tokensinteger

Maximum number of tokens to generate in the response.

presence_penaltyinteger

Penalizes new tokens based on whether they appear in the text so far, encouraging new topics.

frequency_penaltyinteger

Penalizes new tokens based on their frequency in the text so far, reducing repetition.

userstring nullable

A unique identifier for the end-user, useful for tracking and abuse detection.

thread_idstring uuid nullable

ID of an existing thread to continue the conversation.

max_parallel_tool_callsinteger

Maximum number of parallel tool calls (e.g., searches) the agent can make during generation. Higher values may speed up responses, but may reduce on-the-fly adaptivity.

engine'v1' | 'v1.5' | 'v2.0'

The engine to use for the DeepNews agent. 'v1' is deprecated while 'v1.5' is our next-generation research agent with access to more tools, better contextual understanding, and improved reasoning capabilities.'v1.5' streams thinking and tool use using tags <think> </think>, <tool_a> </tool_a> Meanwhile, 'v2.0' is the same underlying agent as 'v1.5', but the thinking and tool use chunks are typed content_block_delta's streamed between content_block_start and content_block_stop events, mimicking Anthropic's approach to streaming tool use/thinking events.

enable_source_pruningboolean

Whether to enable source pruning, which removes sources that are deemed irrelevant to the final response. This can help reduce noise in the sources and improve the quality of the response, but may also remove some relevant sources if not used carefully.

cutoff_datetimestring date-time nullable

The knowledge cutoff datetime for the agent prompt and all tool calls.

stream_sources_inlineboolean

Engine 'v2.0' only: when true, one chat.completion.source.batch chunk per research tool call is streamed after the turn's tool calls stop and before the next agent output, carrying that call's source attribution. Off by default; the final chat.completion.source chunks are unchanged either way.

Example request

{
  "messages": [
    {
      "content": "What are the latest developments in AI regulation?",
      "role": "user"
    }
  ]
}

Response

Deep news research.

idstring required
createdinteger required
objectstring
modelstring

Changes

  • 6513c99e2917Revision changes
    • info

      added subschema #2 subschema #3 to the response body anyOf list for the response status (media type: application/json)

    • info

      removed subschema #2 subschema #3 from the response body anyOf list for the response status (media type: application/json)

    • info

      added the optional property ///////// to the response with the status (media type: application/json)

    • info

      added the optional property //////// to the response with the status (media type: text/event-stream)

    • info

      added the optional property ///////// to the response with the status (media type: text/event-stream)

    • info

      added the optional property //////// to the response with the status (media type: application/json)

  • f1818f4f772bRevision changes
    • breaking

      added subschema #4: CreateDeepNewsResponseStreamSourceBatch to the response body oneOf list for the response status (media type: text/event-stream)

    • breaking

      the response property ///// became optional for the status (media type: application/json)

    • breaking

      the response property ///// became optional for the status (media type: application/json)

    • breaking

      the response property //// became optional for the status (media type: text/event-stream)

    • breaking

      the response property //// became optional for the status (media type: text/event-stream)

    • breaking

      the response property //// became optional for the status (media type: application/json)

    • breaking

      the response property //// became optional for the status (media type: application/json)

    • breaking

      response property anyOf[subschema #1: CreateDeepNewsResponse]/sources/news/anyOf[subschema #1]/items/centroid_distance list-of-types was widened by adding types null to media type application/json of response 200

    • breaking

      response property anyOf[subschema #1: CreateDeepNewsResponse]/sources/news/anyOf[subschema #1]/items/cluster_probability list-of-types was widened by adding types null to media type application/json of response 200

    • breaking

      response property oneOf[subschema #3: CreateDeepNewsResponseStreamSource]/source/anyOf[subschema #1: CreateDeepNewsResponseStreamSourcesNewsSource]/data/centroid_distance list-of-types was widened by adding types null to media type text/event-stream of response 200

    • breaking

      response property oneOf[subschema #3: CreateDeepNewsResponseStreamSource]/source/anyOf[subschema #1: CreateDeepNewsResponseStreamSourcesNewsSource]/data/cluster_probability list-of-types was widened by adding types null to media type text/event-stream of response 200

    • breaking

      response property sources/news/anyOf[subschema #1]/items/centroid_distance list-of-types was widened by adding types null to media type application/json of response 200

    • breaking

      response property sources/news/anyOf[subschema #1]/items/cluster_probability list-of-types was widened by adding types null to media type application/json of response 200

    • info

      added the new optional request property

    • info

      added subschema #2 subschema #3 to the response body anyOf list for the response status (media type: application/json)

    • info

      removed subschema #2 from the response body anyOf list for the response status (media type: application/json)