AI Radar

R1 1776

Retired · Language, Reasoning

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 Sonar API as an offline chat model without search. Perplexity removed it from the API on 1 August 2025; a Llama 70B distilled version was also published. [1] [2] [3] [4] [5] Primary source

Timeline of R1 1776 →

Claims and evidence

  • Released Weights released on Hugging Face and the model offered through the Sonar API, per the post of 2025-02-18; the API model list captured on 2025-02-22 includes r1-1776. Primary source[1]first paragraph [6]Supported Models table
  • Deprecated The notice is labelled August 2025 (month only) and gives the removal date of 2025-08-01. Primary source[4]August 2025: API model deprecation notice
  • Retired Removal from the Perplexity API. Primary source[4]August 2025: API model deprecation notice
  • Status RetiredRemoved from the API on 2025-08-01. The Hugging Face repositories perplexity-ai/r1-1776 and perplexity-ai/r1-1776-distill-llama-70b answered 401 on 2026-10-05 and are absent from the perplexity-ai model listing, so the weights are no longer offered by Perplexity there. Primary source[4]August 2025: API model deprecation notice
  • Replaced by Sonar Reasoning ProThe notice recommends switching to Sonar Pro Reasoning, the changelog's wording for Sonar Reasoning Pro. Primary source[4]August 2025: API model deprecation notice
  • Derived from (fine tune) DeepSeek-R1 (DeepSeek) Archived primaryOriginal source unavailable[1] [2]
    • Open weights YesWeights published on Hugging Face under the MIT licence tag in February 2025; the repository is no longer reachable (401 on 2026-10-05). Archived primaryOriginal source unavailable[1]first paragraph [2]
    • Context window 128k tokensContext length of the API offering. Primary source[6]Supported Models table [7]Context Length
    • Parameters Not disclosedThe model card text states no count. The Hugging Face page widget showed a size computed by Hugging Face from the weight files, not a figure stated by Perplexity. No claim madeNo source
    • Access Open-weights download, APIAs offered in 2025; neither route is available any more. Primary source[1]first paragraph [6]Supported Models table

    Lineage

    Predecessors

    No known predecessor.

    Successors

    No known successor.

    Based on

    • DeepSeek-R1 · DeepSeek · 20 January 2025 · derived (fine tune)

    Variants and derived

    None recorded.

    Siblings

    None recorded.

    All ancestors

    All descendants

    None.

    Variants

    R1 1776 Distill Llama 70B

    • Released Not in the post of 2025-02-18; the repository data gives createdAt 2025-02-21 and the archived card is of 2025-02-22. Public release day not stated. Archived primaryOriginal source unavailable[5]
    • Variant Inline variant in this record.Described by Perplexity as a Llama 70B distilled version of R1 1776; the card's base_model metadata is deepseek-ai/DeepSeek-R1-Distill-Llama-70B. Offered only as weights on Hugging Face, not through the API. Archived primaryOriginal source unavailable[5]

    Also known as: r1-1776-distill-llama-70b, perplexity-ai/r1-1776-distill-llama-70b

    Differs in:

    • Open weights: Yes Evidence not assessed [5]

    Related AI Radar coverage

    AI Radar coverage starts in June 2026; no coverage linked yet.

    All model releases from Perplexity on AI Radar →