Prime Intellect

Prime Intellect is a San Francisco–based startup building an open and decentralized stack for developing and improving AI agents through reinforcement learning. The company combines a 50+ provider GPU compute marketplace (single-node and 1–256 GPU multi-node clusters across H100, H200, B200, and B300), a managed RL post-training service ("Lab") that fine-tunes Qwen3.5, Llama, NVIDIA Nemotron, and OpenAI gpt-oss base models against a 2,500+ environment library (the Environments Hub), secure remote Python sandboxes for agent code execution, OpenAI-compatible inference at api.pinference.ai with LoRA adapter deployments, and an evaluations service that runs against the same environments. Prime Intellect is also the publisher of INTELLECT-3 (100B+ MoE trained with RL), INTELLECT-2 (32B trained via distributed RL), and the SYNTHETIC-2 dataset of four million reasoning traces, and develops the open-source `verifiers` RL library and the `prime-rl` async RL training framework. Everything is exposed via a single bearer-token REST API at api.primeintellect.ai, a `prime` Python CLI/SDK (PyPI: `prime`, lightweight `prime-sandboxes`), and the OpenAI protocol on the inference plane.

6 APIs 0 Features

AIArtificial IntelligenceReinforcement LearningGPU ComputeDecentralized ComputeFoundation ModelsInferenceSandboxesTrainingEnvironmentsEvaluationsLoRAOpen SourceAgents

APIs

Prime Intellect Compute API

Programmatic access to the Prime Intellect compute marketplace. Query GPU availability across 50+ providers, provision and manage on-demand or multi-node GPU pods (1-256 GPUs ac...

Prime Intellect Sandbox API

Secure remote code-execution sandboxes for AI agents. Create and manage Python-3.11-slim-based sandboxes, expose ports for HTTP services running inside the sandbox, open SSH ses...

Prime Intellect Training API

Hosted reinforcement-learning post-training service ("Lab"). Launch dedicated training runs against a catalog of base models — Qwen3.5 (0.8B through 397B-A17B), Llama 1B/3B Inst...

Prime Intellect Evaluations API

Create, manage, and submit samples to evaluations against the Prime Intellect Environments Hub (2,500+ RL environments). Supports both client-driven evaluations (push samples, f...

Prime Intellect Inference API

OpenAI-compatible inference API for hosted frontier and open models served at api.pinference.ai. Supports streaming chat completions, the full set of OpenAI parameters (temperat...

Prime Intellect Platform API

Account, team, secrets, image, and cluster administration. Whoami and team membership, secret CRUD for use inside pods and sandboxes, Docker image builds with public/private vis...

Collections

Prime Intellect Compute API

OPEN

Prime Intellect Evaluations API

OPEN

Prime Intellect Inference API

OPEN

Prime Intellect Platform API

OPEN

Prime Intellect Sandbox API

OPEN

Prime Intellect Training API

OPEN

Pricing Plans

Prime Intellect Plans Pricing

6 plans

PLANS

Rate Limits

Prime Intellect Rate Limits

3 limits

RATE LIMITS

FinOps

Prime Intellect Finops

FINOPS

Semantic Vocabularies

Prime Intellect Context

0 classes · 8 properties

JSON-LD

Resources

Documentation

APIReference

Authentication

HuggingFace

Sources

opencollection: 1.0.0
info:
  name: Prime Intellect Training API
  version: 0.1.0
request:
  auth:
    type: bearer
    token: '{{bearerToken}}'
items:
- info:
    name: Billing
    type: folder
  items:
  - info:
      name: Token usage and cost for a single RFT run
      type: http
    http:
      method: GET
      url: https://api.primeintellect.ai/api/v1/billing/runs/:run_id/usage
      params:
      - name: run_id
        value: ''
        type: path
      auth:
        type: bearer
        token: '{{bearerToken}}'
    docs: 'Return the running token + cost totals for one RFT run.


      Mirrors the per-row training data shown on the billing page (Training

      tokens, Inference input/output tokens, price per million tokens, and

      total cost) so an agent can poll this endpoint to monitor a live run.'
  - info:
      name: Wallet balance + most recent billing rows
      type: http
    http:
      method: GET
      url: https://api.primeintellect.ai/api/v1/billing/wallet
      params:
      - name: limit
        value: ''
        type: query
        description: Number of recent billing rows to include (max 100).
      - name: offset
        value: ''
        type: query
        description: Skip this many rows before returning. Lets callers page.
      - name: teamId
        value: ''
        type: query
        description: If provided, return the team's wallet (requires membership).
      auth:
        type: bearer
        token: '{{bearerToken}}'
    docs: 'Return the wallet''s current balance and most recent billing rows.


      All resource types are included (compute, training, inference, disks,

      sandboxes, images) — same source-of-truth `Billing` table the

      dashboard''s Billing History tab reads. Sorted by `lastBilledAt` desc.'
- info:
    name: training
    type: folder
  items:
  - info:
      name: Create Dedicated Run
      type: http
    http:
      method: POST
      url: https://api.primeintellect.ai/api/v1/training/runs
      body:
        type: json
        data: '{}'
      auth:
        type: bearer
        token: '{{bearerToken}}'
    docs: Dispatch a dedicated full-FT prime-rl run on a registered PrimeCluster.
  - info:
      name: Delete Dedicated Run
      type: http
    http:
      method: DELETE
      url: https://api.primeintellect.ai/api/v1/training/runs/:run_id
      params:
      - name: run_id
        value: ''
        type: path
      auth:
        type: bearer
        token: '{{bearerToken}}'
    docs: 'Delete a dedicated run: helm uninstall + namespace delete + DB cleanup.


      Idempotent. Safe to call repeatedly. Cleans up the linked Job row too

      (which uninstalls the helm release on the customer cluster). Soft-

      deletes the RFTRun row (matches the existing run-delete UX — Billing

      refs are preserved).'
bundled: true