Consortium forVerified Computing

O-0100

Prime Intellect

An AI compute, training and inference company; developer of TOPLOC, a hashing scheme for verifying LLM inference.

Source reviewed 2026-10-08Provider-reported evidence

01 / The mechanism and its boundary

What is being described

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:

  • The ICML 2025 TOPLOC paper describes a compact hash of the largest activations of the last hidden layer S-1000. A verifier who re-runs the sequence compares these hashes to 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.

Organization website ↗

Connections in the research map

Related research

mechanism / R3

Sampled inference recomputation

A verifier re-runs a random sample of an AI provider's logged queries on a trusted copy of the declared model and checks the outputs match.

implementation / R3

TOPLOC

TOPLOC is a hashing scheme from Prime Intellect that lets a verifier check whether an inference provider ran the model, prompt and precision it claims.

Sources and provenance

  1. S-3600 / Tier B

    Prime Intellect homepage ↗

    · 2026 · Prime Intellect

    Supports: integrated compute, training, inference and sandbox stack

    Version and catalogue details
  2. S-1000 / Tier A

    TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference ↗

    J. M. Ong, M. Di Ferrante, A. Pazdera, R. Garner, S. Jaghouar, M. Basra, M. Ryabinin, J. Hagemann · 2025 · Proceedings of the 42nd International Conference on Machine Learning (PMLR 267), pp. 47196-47211

    Supports: TOPLOC paper; seven of the eight authors at Prime Intellect; what the method checks

    Version and catalogue details
  3. S-1001 / Tier B

    PrimeIntellect-ai/toploc (GitHub repository) ↗

    Prime Intellect · 2025 · GitHub

    Supports: open-source TOPLOC package (MIT licence)

    Version and catalogue details
  4. S-1002 / Tier C

    TOPLOC: A Locality Sensitive Hashing Scheme for Trustless Verifiable Inference (blog post) ↗

    Prime Intellect · 2025 · Prime Intellect blog

    Supports: vLLM and SGLang integrations (provider-reported)

    Version and catalogue details
  5. S-1003 / Tier B

    INTELLECT-2: A Reasoning Model Trained Through Globally Decentralized Reinforcement Learning ↗

    Prime Intellect Team, S. Jaghouar, J. Mattern, J. M. Ong, J. Straube, M. Basra, A. Pazdera, K. Thaman, M. Di Ferrante, F. Gabriel, F. Obeid, K. Erdem, M. Keiblinger, J. Hagemann · 2025 · arXiv

    Supports: TOPLOC proofs in the INTELLECT-2 training run (provider-reported)

    Version and catalogue details
  6. S-1004 / Tier C

    SYNTHETIC-2 ↗

    Prime Intellect · 2025 · Prime Intellect blog

    Supports: TOPLOC v2 extends the scheme to pipeline-parallel inference and sampling checks (provider-reported)

    Version and catalogue details
  7. S-3000 / Tier C

    SYNTHETIC-2 Release: Four Million Collaboratively Generated Reasoning Traces ↗

    Prime Intellect · 2025 · Prime Intellect blog

    Supports: TOPLOC v2 checks in the SYNTHETIC-2 run: 1,250+ GPUs, false-positive rate over 4 million samples (provider-reported)

    Version and catalogue details
Source review date
2026-10-08
Drafted by (source map)
ai
Review handles (source map)
codex-review