๐Ÿฆ Twitter Post Details

Viewing enriched Twitter post

@LiorOnAI

Small models just beat giant LLM agents at their own job. Not by thinking harder, but by coordinating better. A new system just outscored GPT-5 on Humanityโ€™s Last Exam, using far less compute. ๐—ง๐—ต๐—ถ๐˜€ ๐˜€๐˜†๐˜€๐˜๐—ฒ๐—บ ๐—ฟ๐—ฒ๐—ฝ๐—น๐—ฎ๐—ฐ๐—ฒ๐˜€ ๐—ผ๐—ป๐—ฒ ๐—ฏ๐—ถ๐—ด ๐—ฏ๐—ฟ๐—ฎ๐—ถ๐—ป ๐˜„๐—ถ๐˜๐—ต ๐—ฎ ๐—ฐ๐—ผ๐—ป๐—ฑ๐˜‚๐—ฐ๐˜๐—ผ๐—ฟ Instead of one model doing everything, it assigns roles. โ€ข Large models handle hard reasoning. โ€ข Small models handle routine steps. โ€ข A controller decides what to call, when. That controller is trained only to make decisions. ๐—œ๐˜ ๐—น๐—ฒ๐—ฎ๐—ฟ๐—ป๐˜€ ๐—ฐ๐—ผ๐—ผ๐—ฟ๐—ฑ๐—ถ๐—ป๐—ฎ๐˜๐—ถ๐—ผ๐—ป, ๐—ป๐—ผ๐˜ ๐—ฝ๐—ฟ๐—ผ๐—บ๐—ฝ๐˜ ๐˜๐—ฟ๐—ถ๐—ฐ๐—ธ๐˜€ It uses reinforcement learning, not hand rules. Rewards optimize three things at once: - Task success - Latency - Compute cost ๐—ง๐—ต๐—ฒ ๐—ฟ๐—ฒ๐˜€๐˜‚๐—น๐˜๐˜€ It scores 37.1% on HLE versus 35.1%. Runs about 2.5ร— faster. Uses roughly 70% less cost. This lets you build agents that scale by coordination, not parameters.

Media 1

๐Ÿ“Š Media Metadata

{
  "media": [
    {
      "type": "photo",
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/2012918090107875549/media_0.jpg?",
      "filename": "media_0.jpg"
    }
  ],
  "processed_at": "2026-01-18T17:32:26.818117",
  "pipeline_version": "2.0"
}

๐Ÿ”ง Raw API Response

{
  "type": "tweet",
  "id": "2012918090107875549",
  "url": "https://x.com/LiorOnAI/status/2012918090107875549",
  "twitterUrl": "https://twitter.com/LiorOnAI/status/2012918090107875549",
  "text": "Small models just beat giant LLM agents at their own job. \n\nNot by thinking harder, but by coordinating better.\n\nA new system just outscored GPT-5 on Humanityโ€™s Last Exam, using far less compute.\n\n๐—ง๐—ต๐—ถ๐˜€ ๐˜€๐˜†๐˜€๐˜๐—ฒ๐—บ ๐—ฟ๐—ฒ๐—ฝ๐—น๐—ฎ๐—ฐ๐—ฒ๐˜€ ๐—ผ๐—ป๐—ฒ ๐—ฏ๐—ถ๐—ด ๐—ฏ๐—ฟ๐—ฎ๐—ถ๐—ป ๐˜„๐—ถ๐˜๐—ต ๐—ฎ ๐—ฐ๐—ผ๐—ป๐—ฑ๐˜‚๐—ฐ๐˜๐—ผ๐—ฟ\nInstead of one model doing everything, it assigns roles.\nโ€ข Large models handle hard reasoning.\nโ€ข Small models handle routine steps.\nโ€ข A controller decides what to call, when.\n\nThat controller is trained only to make decisions.\n\n๐—œ๐˜ ๐—น๐—ฒ๐—ฎ๐—ฟ๐—ป๐˜€ ๐—ฐ๐—ผ๐—ผ๐—ฟ๐—ฑ๐—ถ๐—ป๐—ฎ๐˜๐—ถ๐—ผ๐—ป, ๐—ป๐—ผ๐˜ ๐—ฝ๐—ฟ๐—ผ๐—บ๐—ฝ๐˜ ๐˜๐—ฟ๐—ถ๐—ฐ๐—ธ๐˜€\nIt uses reinforcement learning, not hand rules.\n\nRewards optimize three things at once:\n- Task success\n- Latency\n- Compute cost\n\n๐—ง๐—ต๐—ฒ ๐—ฟ๐—ฒ๐˜€๐˜‚๐—น๐˜๐˜€ \nIt scores 37.1% on HLE versus 35.1%.\nRuns about 2.5ร— faster.\nUses roughly 70% less cost.\n\nThis lets you build agents that scale by coordination, not parameters.",
  "source": "Twitter for iPhone",
  "retweetCount": 0,
  "replyCount": 4,
  "likeCount": 8,
  "quoteCount": 0,
  "viewCount": 1045,
  "createdAt": "Sun Jan 18 16:00:45 +0000 2026",
  "lang": "en",
  "bookmarkCount": 4,
  "isReply": false,
  "inReplyToId": null,
  "conversationId": "2012918090107875549",
  "displayTextRange": [
    0,
    263
  ],
  "inReplyToUserId": null,
  "inReplyToUsername": null,
  "author": {
    "type": "user",
    "userName": "LiorOnAI",
    "url": "https://x.com/LiorOnAI",
    "twitterUrl": "https://twitter.com/LiorOnAI",
    "id": "931470139",
    "name": "Lior Alexander",
    "isVerified": false,
    "isBlueVerified": true,
    "verifiedType": null,
    "profilePicture": "https://pbs.twimg.com/profile_images/1980366446975766528/LPbXxZYl_normal.jpg",
    "coverPicture": "https://pbs.twimg.com/profile_banners/931470139/1761077189",
    "description": "",
    "location": "",
    "followers": 109076,
    "following": 1982,
    "status": "",
    "canDm": true,
    "canMediaTag": false,
    "createdAt": "Wed Nov 07 07:19:36 +0000 2012",
    "entities": {
      "description": {
        "urls": []
      },
      "url": {}
    },
    "fastFollowersCount": 0,
    "favouritesCount": 6604,
    "hasCustomTimelines": true,
    "isTranslator": false,
    "mediaCount": 638,
    "statusesCount": 3632,
    "withheldInCountries": [],
    "affiliatesHighlightedLabel": {},
    "possiblySensitive": false,
    "pinnedTweetIds": [],
    "profile_bio": {
      "description": "Daily posts on the latest dev news in AI โ€ข Founder @AlphaSignalAI (250k users) โ€ข ML Eng since 2017 โ€ข Ex-Mila",
      "entities": {
        "description": {
          "user_mentions": [
            {
              "id_str": "0",
              "indices": [
                51,
                65
              ],
              "name": "",
              "screen_name": "AlphaSignalAI"
            }
          ]
        },
        "url": {
          "urls": [
            {
              "display_url": "alphasignal.ai",
              "expanded_url": "https://alphasignal.ai",
              "indices": [
                0,
                23
              ],
              "url": "https://t.co/AyubevadmD"
            }
          ]
        }
      }
    },
    "isAutomated": false,
    "automatedBy": null
  },
  "extendedEntities": {
    "media": [
      {
        "display_url": "pic.twitter.com/R5eAGWcoQo",
        "expanded_url": "https://twitter.com/LiorOnAI/status/2012918090107875549/photo/1",
        "ext_media_availability": {
          "status": "Available"
        },
        "features": {
          "large": {
            "faces": [
              {
                "h": 125,
                "w": 125,
                "x": 1,
                "y": 389
              }
            ]
          },
          "orig": {
            "faces": [
              {
                "h": 179,
                "w": 179,
                "x": 2,
                "y": 554
              }
            ]
          }
        },
        "id_str": "2012918087301935105",
        "indices": [
          264,
          287
        ],
        "media_key": "3_2012918087301935105",
        "media_results": {
          "id": "QXBpTWVkaWFSZXN1bHRzOgwAAQoAARvvUlXjlzABCgACG+9SVorWcN0AAA==",
          "result": {
            "__typename": "ApiMedia",
            "id": "QXBpTWVkaWE6DAABCgABG+9SVeOXMAEKAAIb71JWitZw3QAA",
            "media_key": "3_2012918087301935105"
          }
        },
        "media_url_https": "https://pbs.twimg.com/media/G-9SVeOXMAEPWHI.jpg",
        "original_info": {
          "focus_rects": [
            {
              "h": 914,
              "w": 1632,
              "x": 0,
              "y": 0
            },
            {
              "h": 914,
              "w": 914,
              "x": 0,
              "y": 0
            },
            {
              "h": 914,
              "w": 802,
              "x": 0,
              "y": 0
            },
            {
              "h": 914,
              "w": 457,
              "x": 136,
              "y": 0
            },
            {
              "h": 914,
              "w": 2914,
              "x": 0,
              "y": 0
            }
          ],
          "height": 914,
          "width": 2914
        },
        "sizes": {
          "large": {
            "h": 642,
            "w": 2048
          }
        },
        "type": "photo",
        "url": "https://t.co/R5eAGWcoQo"
      }
    ]
  },
  "card": null,
  "place": {},
  "entities": {},
  "quoted_tweet": null,
  "retweeted_tweet": null,
  "isLimitedReply": false,
  "article": null
}