AI Radar

Perplexity models

Perplexity offers an answer engine that gives cited answers, Perplexity Computer for multi-step research and workflow tasks, the Comet AI browser and an API platform for developers. It describes routing work across models from several providers and calls Sonar its in-house model, optimised for answer quality in its default search mode.

Chronology

  1. 2026

    1 model
    1. Released
      PPLX · Language PPLX 27B

      PPLX 27B is a language model that Perplexity post-trained from Qwen 3.8 27B, introduced on 25 August 2026 with Portable Computer, a version of Perplexity Computer that runs entirely on a local…

      Primary source Available · no news linked

  2. 2025

    7 models
    1. Released
      Sonar · Language, Reasoning R1 1776

      R1 1776 is a version of DeepSeek-R1 that Perplexity post-trained to remove Chinese Communist Party censorship, published as open weights under the MIT licence on 18 February 2025 and offered in the…

      Primary source Retired · no news linked
    2. Released
      Sonar · Language Sonar (February 2025)

      The February 2025 version of Sonar is Perplexity's in-house model built on Llama 3.3 70B, released under the unchanged name Sonar for Perplexity Pro users in the app and also usable through the…

      • Built on Llama 3.3 70B and further trained for answer factuality and readability in Perplexity's default search mode.
      • Offered to Perplexity Pro users in the app, who could set it as their default model, in addition to the Sonar API.
      Primary source Deprecated · no news linked
    3. Day not given
      Sonar · Language, Reasoning Sonar Deep Research

      Sonar Deep Research is the research model of Perplexity's Sonar API, first listed in the API docs in late February 2025 and offered only through the API. Perplexity's docs say it runs dozens of…

      Primary source Deprecated · no news linked
    4. Uncertain month
      Sonar Reasoning · Language, Reasoning Sonar Reasoning Pro

      Sonar Reasoning Pro is a reasoning model in Perplexity's Sonar API, first listed in the API docs in early February 2025, that returns its chain of thought in the response. Perplexity's docs say it…

      • Perplexity's docs describe it as powered by DeepSeek R1 with chain of thought.
      • Has an 8k maximum output, a limit the docs state for this model and Sonar Pro but not for Sonar Reasoning.
      Primary source Deprecated · no news linked
    5. Released
      Sonar · Language · Milestone Sonar

      Sonar is the lightweight model of Perplexity's Sonar API, launched on 21 January 2025 together with Sonar Pro and offered only through the API. Perplexity aimed it at companies that need quick…

      • Comes with citations included and source customisation; Perplexity also opened JSON mode and search domain filters to more usage tiers.
      Primary source Deprecated · no news linked
    6. Released
      Sonar · Language, Multimodal Sonar Pro

      Sonar Pro is the higher tier of Perplexity's Sonar API, made generally available to developers on 21 January 2025 alongside Sonar. Perplexity positions it for enterprises with in-depth, multi-step…

      • Context length of 200k tokens, up from 127k for the llama-3.1-sonar online models it replaced.
      • Perplexity says it returns on average twice as many citations per search as Sonar and is meant for multi-step queries.
      Primary source Deprecated · no news linked
    7. Day not given
      Sonar Reasoning · Language, Reasoning Sonar Reasoning

      Sonar Reasoning was a reasoning model in Perplexity's Sonar API, first listed in the API docs at the end of January 2025, that returned its chain of thought in the response. It had a 127k context…

      • Adds chain-of-thought reasoning, returned in the response, on top of Sonar's web search with citations.
      Primary source Retired · no news linked

  3. 2024

    3 models
    1. Day not given
      Sonar · Language Llama 3.1 Sonar

      Llama 3.1 Sonar covers Perplexity API language models launched in July 2024: llama-3.1-sonar-small-128k and llama-3.1-sonar-large-128k as online (search) and chat ids, joined by August 2024 by an…

      • Context length grew to 131,072 tokens for the chat ids and 127,072 for the online ids, from 32768 and 28000.
      • Adds a third, online-only model, llama-3.1-sonar-huge-128k, listed at 405B parameters; it was on the API docs by August 2024.
      Primary source Retired · no news linked
    2. Uncertain month
      Sonar · Language Llama 3 Sonar

      Llama 3 Sonar covers two Perplexity API language models, llama-3-sonar-small-32k and llama-3-sonar-large-32k, each offered as a chat id and an online id with search, listed at 7B and 70B parameters…

      • Context length grew to 32768 tokens for the chat ids and 28000 for the online ids, from 16384 and 12000.
      • The larger tier is listed at 70B parameters, where sonar-medium was listed at 8x7B.
      Archived primary Retired · no news linked Original source unavailable
    3. Released
      Sonar · Language · Milestone Sonar (February 2024)

      Sonar (February 2024) refers to Perplexity's sonar-small and sonar-medium language models, launched on 23 February 2024 on the Perplexity API, each as a chat id and a search-enhanced online id. The…

      • Context length of 16384 tokens for the chat ids and 12000 for the online ids, up from 4096 for the PPLX online models at release.
      • The larger tier, sonar-medium, is listed at 8x7B parameters in mixture-of-experts notation; the larger PPLX online model was pplx-70b-online.
      Primary source Retired · no news linked

  4. 2023

    1 model
    1. Released
      PPLX · Language · Milestone PPLX Online LLMs

      PPLX Online LLMs were two Perplexity language models, pplx-7b-online and pplx-70b-online, released on 29 November 2023 through pplx-api and the Perplexity Labs playground. Perplexity says they were…

      Primary source Retired · no news linked

Lineage

  1. PPLX Online LLMs
    Based on
    Mistral 7B (Mistral AI) · derived: fine tune; Llama 2 (Meta) · derived: fine tune
  2. Sonar (February 2024)
    Predecessors
    PPLX Online LLMs
    Successors
    Llama 3 Sonar
  3. Llama 3 Sonar
    Successors
    Llama 3.1 Sonar
  4. Llama 3.1 Sonar
    Predecessors
    Llama 3 Sonar
    Successors
    Sonar; Sonar Pro
  5. Sonar Reasoning

    No recorded relations.

  6. Sonar
    Predecessors
    Llama 3.1 Sonar
    Variants and derived
    Sonar (February 2025) · revision
  7. Sonar Pro
    Predecessors
    Llama 3.1 Sonar
  8. Sonar Deep Research

    No recorded relations.

  9. Sonar Reasoning Pro
    Based on
    DeepSeek-R1 (DeepSeek) · derived: other
  10. Sonar (February 2025)
    Based on
    Sonar · revision; Llama 3.3 (Meta) · derived: fine tune
  11. R1 1776
    Based on
    DeepSeek-R1 (DeepSeek) · derived: fine tune
  12. PPLX 27B

    No recorded relations.

Families

About the organisation

  • Description Perplexity offers an answer engine that gives cited answers, Perplexity Computer for multi-step research and workflow tasks, the Comet AI browser and an API platform for developers. It describes routing work across models from several providers and calls Sonar its in-house model, optimised for answer quality in its default search mode. Primary source[1] [2]

Company · https://www.perplexity.ai · 12 models listed here

Sources

  1. [1] Perplexity | AI for the Curious

    Other · Perplexity

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://www.perplexity.ai/hub Internet Archive capture ↗ captured Company overview page. perplexity.ai answers 403 to scripted requests, so the health check is skipped; content was read via the Internet Archive capture of 2026-09-02.

  2. [2] Meet New Sonar: A blazing fast model optimized for Perplexity search

    Announcement · Perplexity · published

    Provenance
    Primary
    Availability
    Active
    Last checked
    Original ↗ https://www.perplexity.ai/hub/blog/meet-new-sonar Internet Archive capture ↗ captured Title from the page heading; the HTML title reads Meet New Sonar: A Blazing Fast Model Optimized for Perplexity Search. Page shows Feb 11, 2025. perplexity.ai answers automated requests with 403, so the content was read from the Internet Archive capture of 2026-09-20.