2025 Changelog ​

November 2025

September 2025

  • NACE Code Rev 2.1 Updates

    KG DATA CHANGE NOTIFICATION - Organization.naceClassification

    We will be updating Organization.naceClassification to NACE Rev. 2.1 in build v437 of the Diffbot Knowledge Graph, targeted to go live in about two weeks. Please read on for more details.

    Ordinarily, we take extraordinary measures to avoid breaking changes in the Diffbot Knowledge Graph ontologies. However, in some cases, there is no benefit in retaining a prior version of the data, so we replace an existing attribute with a new data format. The Organization.naceClassification field is one such case. The current version of the NACE codes in the KG lacks level, isPrimary, and ancestor codes. And, some of the codes are no longer valid in the latest NACE Rev. 2.1 version.

    In Rev 2.1 of the NACE codes:

    • NACE codes are no longer strictly 4-digit numbers.
    • NACE codes are structured into:
          Sections (letters A–V, level 1) →
 Divisions (2 digits, level 2) →
          Groups (3 digits with dot, level 3) →
 Classes (4 digits with dot, level 4).
    • Codes are unique. For example, both 28 and 29 share the same parent C, but it is not repeated after 28 because it already appears earlier in the primary chain.
    • There is at most one primary code per level.
    • Primary codes are listed before non-primaries.
    • Specific codes (e.g., 29.10) are listed before broader ones (e.g., 29.1).

    For a comparison of the existing code format versus the new Rev 2.1 format, see below.

    CURRENT DATA FORMAT: NACE codes - Organization.naceClassification

    Volkswagen's current NACE classification in the KG appears as the following

    json
    [  
  {  
    "code": "2910",  
    "isPrimary": false,  
    "name": "Manufacture of motor vehicles"  
  },  
  {  
    "code": "7022",  
    "isPrimary": false,  
    "name": "Business and other management consultancy activities"  
  },  
  {  
    "code": "7021",  
    "isPrimary": false,  
    "name": "Public relations and communication activities"  
  }  
]

    Issues with this data:

    • Missing level information
    • All codes are marked as non-primary
    • Parent codes are missing
    • Codes 7022 and 7021 are no longer valid in the Rev. 2.1 version of the codes
    • Volkswagen should not be classified under those industries in 7022 and 7021.

    NEW DATA FORMAT: NACE Rev 2.1 Codes

    When the updates deploy, Volkswagen's Organization.naceClassification NACE codes will look like this:

    json
    [  
  {  
    "code": "29.10",  
    "level": 4,  
    "isPrimary": true,  
    "name": "Manufacture of motor vehicles",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "29.1",  
    "level": 3,  
    "isPrimary": true,  
    "name": "Manufacture of motor vehicles",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "29",  
    "level": 2,  
    "isPrimary": true,  
    "name": "Manufacture of motor vehicles, trailers and semi-trailers",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "C",  
    "level": 1,  
    "isPrimary": true,  
    "name": "MANUFACTURING",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "28.11",  
    "level": 4,  
    "isPrimary": false,  
    "name": "Manufacture of engines and turbines, except aircraft, vehicle and cycle engines",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "28.1",  
    "level": 3,  
    "isPrimary": false,  
    "name": "Manufacture of general-purpose machinery",  
    "version": "Rev 2.1"  
  },  
  {  
    "code": "28",  
    "level": 2,  
    "isPrimary": false,  
    "name": "Manufacture of machinery and equipment n.e.c.",  
    "version": "Rev 2.1"  
  }  
]

  • Diffbot LLM live on LMArena

    diffbot-small-xl is finally live on https://lmarena.ai/! Check out: https://lmarena.ai/leaderboard/search. LMArena has open-sourced the largest repository of organic human preferences on generative models in the world. These datasets are free and open to access.

June 2025

May 2025

April 2025

  • Invite a User

    We have updated the dashboard to better support managing your team. You can now invite a user to share your primary (parent token) to manage shared bulk extraction tasks, bulk enrichment tasks, and crawl jobs. Or, designate a child token for that user so that they access Diffbot services independently using your account budget. Check out the new features in your Dashboard UI here .

  • New DQL JSON Export filter - ' &filterExclude'

    When exporting data from collections via DQL, you have always had the option of specifying ONLY the fields you want to be returned in the JSON output by using the '&filter=' param (i.e. &filter=id%20name%20homepageUri added to a query like this).

    https://kg.diffbot.com/kg/v3/dql?type=query&token=TOKEN&query=type%3AOrganization+types%3A%22Company%22&size=25&filter=id%20name%20homepageUri

    But this approach can be unwieldy if you have a long list of attributes to include or if you only want to exclude a few attributes per entity in the output.

    Now, instead of specifying the fields you want to include, you can exclude fields you do not want returned when exporting data by using the &filterExclude= param (i.e. &filterExclude=subsidiaries%20technologies%20customers)

    https://kg.diffbot.com/kg/v3/dql?type=query&token=TOKEN&query=type%3AOrganization+types%3A%22Company%22&size=25&filterExclude=subsidiaries%20technologies%20customers

  • PDF Invoices

    We added PDF exports to invoices that match your monthly billing periods (including any overage fees owed/paid). Try out the new Billing features here .

    Usage Analytics are still available via the Account API and via the Diffbot Dashboard.

March 2025

January 2025

  • Diffbot GraphRAG LLM

    Recently, large language models (LLMs) have been trained with more and more data, leading to an increase in the number of parameters and the computing power needed. But, what if, instead of feeding the model more data, we purposefully trained it to rely less on its pretraining data and more on its ability to find external knowledge?

    To test this idea, we fine-tuned LLama 3.3 70B to be an expert tool user of a real-time Knowledge Graph API, providing the first open-source implementation of a GraphRAG system that outperforms Google Gemini and ChatGPT. To learn more, see: https://github.com/diffbot/diffbot-llm-inference/.