{
  "schema_version": "1.2.0",
  "rubric_version": "1.1",
  "license": "CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)",
  "record": {
    "id": "O-0100",
    "slug": "prime-intellect",
    "title": "Prime Intellect",
    "aliases": [],
    "status": "published",
    "last_reviewed": "2026-09-25",
    "review_interval_days": 90,
    "steward": null,
    "provenance": {
      "drafted_by": "ai",
      "reviewed_by": [
        "codex-review"
      ]
    },
    "risk_flags": [],
    "flags": [
      "provider-reported"
    ],
    "kind": "company",
    "homepage": "https://www.primeintellect.ai/",
    "one_liner": "An AI compute, training and inference company; developer of TOPLOC, a hashing scheme for verifying LLM inference.",
    "sources": [
      {
        "source": "S-3600",
        "supports": "integrated compute, training, inference and sandbox stack"
      },
      {
        "source": "S-1000",
        "supports": "TOPLOC paper; seven of the eight authors at Prime Intellect; what the method checks"
      },
      {
        "source": "S-1001",
        "supports": "open-source TOPLOC package (MIT licence)"
      },
      {
        "source": "S-1002",
        "supports": "vLLM and SGLang integrations (provider-reported)"
      },
      {
        "source": "S-1003",
        "supports": "TOPLOC proofs in the INTELLECT-2 training run (provider-reported)"
      },
      {
        "source": "S-1004",
        "supports": "TOPLOC v2 extends the scheme to pipeline-parallel inference and sampling checks (provider-reported)"
      },
      {
        "source": "S-3000",
        "supports": "TOPLOC v2 checks in the SYNTHETIC-2 run: 1,250+ GPUs, false-positive rate over 4 million samples (provider-reported)"
      }
    ],
    "type": "organization",
    "url": "https://trustbutveri.fyi/organizations/prime-intellect/",
    "source_file": "content/organizations/prime-intellect.md",
    "flags_all": [
      "provider-reported"
    ],
    "body_markdown": "Prime Intellect describes itself as offering an integrated compute, training, inference and sandbox stack for AI models [[S-3600]]. Its verification work centres on [[I-0001|TOPLOC]], a form of [[M-0001|sampled inference recomputation]]:\n\n- Seven of the eight authors of the ICML 2025 paper that introduces TOPLOC are at Prime Intellect [[S-1000]]. An inference provider using it records the largest activations of the last hidden layer, so that a verifier who re-runs the sequence can check the claimed model, prompt and precision [[S-1000]].\n- Prime Intellect publishes TOPLOC as an MIT-licensed Python package [[S-1001]], and reports integrations with vLLM and a fork of SGLang [[S-1002]].\n- Prime Intellect reports that each inference worker in its INTELLECT-2 decentralized training run generated a TOPLOC proof for every sequence [[S-1003]]. It reports that a later version, TOPLOC v2, extends the scheme to pipeline-parallel inference and adds checks of token sampling [[S-1004]]. It reports using TOPLOC v2 to check the inference workers in its SYNTHETIC-2 data-generation run, on more than 1,250 GPUs, with a false-positive rate of 0.000925% over 4 million samples [[S-3000]].",
    "body_text": "Prime Intellect describes itself as offering an integrated compute, training, inference and sandbox stack for AI models [S-3600]. Its verification work centres on TOPLOC, a form of sampled inference recomputation: - Seven of the eight authors of the ICML 2025 paper that introduces TOPLOC are at Prime Intellect [S-1000]. An inference provider using it records the largest activations of the last hidden layer, so that a verifier who re-runs the sequence can check the claimed model, prompt and precision [S-1000]. - Prime Intellect publishes TOPLOC as an MIT-licensed Python package [S-1001], and reports integrations with vLLM and a fork of SGLang [S-1002]. - Prime Intellect reports that each inference worker in its INTELLECT-2 decentralized training run generated a TOPLOC proof for every sequence [S-1003]. It reports that a later version, TOPLOC v2, extends the scheme to pipeline-parallel inference and adds checks of token sampling [S-1004]. It reports using TOPLOC v2 to check the inference workers in its SYNTHETIC-2 data-generation run, on more than 1,250 GPUs, with a false-positive rate of 0.000925% over 4 million samples [S-3000].",
    "referenced_by": [
      {
        "id": "M-0001",
        "title": "Sampled inference recomputation",
        "url": "https://trustbutveri.fyi/mechanisms/sampled-inference-recomputation/"
      },
      {
        "id": "I-0001",
        "title": "TOPLOC",
        "url": "https://trustbutveri.fyi/implementations/toploc/"
      },
      {
        "id": "S-1000",
        "title": "TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference",
        "url": "https://trustbutveri.fyi/sources/ong-toploc/"
      },
      {
        "id": "S-3600",
        "title": "Prime Intellect homepage",
        "url": "https://trustbutveri.fyi/sources/prime-intellect-homepage/"
      },
      {
        "id": "S-1003",
        "title": "INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning",
        "url": "https://trustbutveri.fyi/sources/primeintellect-intellect-2/"
      },
      {
        "id": "S-3000",
        "title": "SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces",
        "url": "https://trustbutveri.fyi/sources/primeintellect-synthetic-2-release/"
      },
      {
        "id": "S-1004",
        "title": "SYNTHETIC-2",
        "url": "https://trustbutveri.fyi/sources/primeintellect-synthetic-2/"
      },
      {
        "id": "S-1002",
        "title": "TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference (blog post)",
        "url": "https://trustbutveri.fyi/sources/primeintellect-toploc-blog/"
      },
      {
        "id": "S-1001",
        "title": "PrimeIntellect-ai/toploc (GitHub repository)",
        "url": "https://trustbutveri.fyi/sources/primeintellect-toploc-code/"
      }
    ]
  }
}