🐦 Twitter Post Details

Viewing enriched Twitter post

@CaimingXiong

Meet SFR-DeepResearch (SFR-DR) πŸ€–: our RL-trained autonomous agents that can reason, search, and code their way through deep research tasks. πŸš€SFR-DR-20B achieves 28.7% on Humanity's Last Exam (text-only) using only web search πŸ”, browsing 🌐, and Python interpreter 🐍, surpassing DeepResearch with OpenAI o3 and Kimi Researcher. πŸ€–SFR-DR agents are trained to operate independently, without pre-defined multi-agent workflows. They autonomously plan, reason, and propose and take actions as defined by their tools. πŸ”„SFR-DR agents are trained with end-to-end RL. Starting from reasoning optimized models, our RL pipeline carefully preserves reasoning abilities while training models to become more capable research agents. πŸ“SFR-DR agents are also trained to manage their own memory by summarizing previous results when context becomes limited. This enables a virtually unlimited context window, enabling long-horizon tasks Paper: https://t.co/32idhdknhh #AIAgents #ReinforcementLearning #DeepResearch

Media 1

πŸ“Š Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1965440617334685886/media_0.jpg?",
      "media_url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1965440617334685886/media_0.jpg?",
      "type": "photo",
      "filename": "media_0.jpg"
    }
  ],
  "processed_at": "2025-09-18T13:50:51.514246",
  "pipeline_version": "2.0"
}

πŸ”§ Raw API Response

{
  "type": "tweet",
  "id": "1965440617334685886",
  "url": "https://x.com/CaimingXiong/status/1965440617334685886",
  "twitterUrl": "https://twitter.com/CaimingXiong/status/1965440617334685886",
  "text": "Meet SFR-DeepResearch (SFR-DR) πŸ€–: our RL-trained autonomous agents that can reason, search, and code their way through deep research tasks.\n\nπŸš€SFR-DR-20B achieves 28.7% on Humanity's Last Exam (text-only) using only web search πŸ”, browsing 🌐, and Python interpreter 🐍, surpassing DeepResearch with OpenAI o3 and Kimi Researcher.\n\nπŸ€–SFR-DR agents are trained to operate independently, without pre-defined multi-agent workflows. They autonomously plan, reason, and propose and take actions as defined by their tools.\n\nπŸ”„SFR-DR agents are trained with end-to-end RL. Starting from reasoning optimized models, our RL pipeline carefully preserves reasoning abilities while training models to become more capable research agents.\n\nπŸ“SFR-DR agents are also trained to manage their own memory by summarizing previous results when context becomes limited. This enables a virtually unlimited context window, enabling long-horizon tasks\n\nPaper: https://t.co/32idhdknhh\n#AIAgents #ReinforcementLearning #DeepResearch",
  "source": "Twitter for iPhone",
  "retweetCount": 145,
  "replyCount": 28,
  "likeCount": 949,
  "quoteCount": 25,
  "viewCount": 297582,
  "createdAt": "Tue Sep 09 15:42:13 +0000 2025",
  "lang": "en",
  "bookmarkCount": 703,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "1965440617334685886",
  "displayTextRange": [
    0,
    267
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "CaimingXiong",
    "url": "https://x.com/CaimingXiong",
    "twitterUrl": "https://twitter.com/CaimingXiong",
    "id": "325839895",
    "name": "Caiming Xiong",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/620325010444423168/rPerKdzU_normal.jpg",
    "coverPicture": "",
    "description": "",
    "location": "Salesforce",
    "followers": 7345,
    "following": 466,
    "status": "",
    "canDm": false,
    "canMediaTag": true,
    "createdAt": "Tue Jun 28 23:30:21 +0000 2011",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 1209,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 203,
    "statusesCount": 1044,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [],
    "profile_bio": {
      "description": "SVP, AI Research Lead at @Salesforce | ex-MetaMind (Opinions are personal.)",
      "entities": {
        "description": {
          "user_mentions": [
            {
              "id_str": "0",
              "indices": [
                25,
                36
              ],
              "name": "",
              "screen_name": "Salesforce"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "scholar.google.com/citations?user…",
              "expanded_url": "https://scholar.google.com/citations?user=vaSdahkAAAAJ&hl=en",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/mtRYscjROq"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.twitter.com/k821mR9ZE6",
        "expanded_url": "https://twitter.com/CaimingXiong/status/1965440617334685886/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {},
          "orig": {}
        },
        "id_str": "1965439459052453889",
        "indices": [
          268,
          291
        ],
        "media_key": "3_1965439459052453889",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARtGpMZQW4ABCgACG0al0/9bwL4AAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABG0akxlBbgAEKAAIbRqXT/1vAvgAA",
            "media_key": "3_1965439459052453889"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/G0akxlBbgAEZNaT.jpg",
        "original_info": {
          "focus_rects": [
            {
              "h": 890,
              "w": 1590,
              "x": 0,
              "y": 138
            },
            {
              "h": 1028,
              "w": 1028,
              "x": 0,
              "y": 0
            },
            {
              "h": 1028,
              "w": 902,
              "x": 0,
              "y": 0
            },
            {
              "h": 1028,
              "w": 514,
              "x": 21,
              "y": 0
            },
            {
              "h": 1028,
              "w": 1590,
              "x": 0,
              "y": 0
            }
          ],
          "height": 1028,
          "width": 1590
        },
        "sizes": {
          "large": {
            "h": 1028,
            "w": 1590
          }
        },
        "type": "photo",
        "url": "https://t.co/k821mR9ZE6"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {
    "hashtags": [
      {
        "indices": [
          953,
          962
        ],
        "text": "AIAgents"
      },
      {
        "indices": [
          963,
          985
        ],
        "text": "ReinforcementLearning"
      },
      {
        "indices": [
          986,
          999
        ],
        "text": "DeepResearch"
      }
    ],
    "urls": [
      {
        "display_url": "arxiv.org/abs/2509.06283",
        "expanded_url": "https://arxiv.org/abs/2509.06283",
        "indices": [
          929,
          952
        ],
        "url": "https://t.co/32idhdknhh"
      }
    ]
  },
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "article": null
}