Prime Intellect
Infrastructure for training, evaluating and running AI agents. Lab brings reinforcement learning environments, hosted training and evaluations, and Prime Intellect adds secure sandboxes, an inference API and GPUs from 50+ providers.
Prime Intellect provides infrastructure for training, evaluating and running AI agents. Its Lab platform brings together the Environments Hub, Hosted Training and Hosted Evaluations for post-training research, and alongside it sit Prime Sandboxes for running agent code, an inference API, and GPU compute from more than fifty providers.
Lab is built around environments. Each one holds a dataset of tasks, the harness the model runs in (tools, sandboxes, context and multi-turn interaction) and a rubric of reward functions and metrics that scores it. The Environments Hub is a community registry of environments for math, coding, games, search and tool use, with validated benchmarks such as AIME and Humanity's Last Exam.
Hosted Evaluations run any environment against Prime Inference or the customer's own OpenAI-compatible endpoint and store the results. Hosted Training runs reinforcement learning with LoRA on open-weights models such as Qwen, Llama and gpt-oss, priced per million tokens, and trained adapters deploy behind an OpenAI-compatible API.
Prime Sandboxes are hardware-isolated microVMs for AI-assisted coding, benchmarks and agent code, with network allow and deny lists, encrypted secrets and custom images, billed per resource hour. GPU instances can be filtered by region and data center before provisioning, and multi-node clusters run with Slurm. The verifiers and prime-rl libraries are open source and run on the customer's own hardware.
Pricing is usage based from prepaid credits: GPUs per hour by provider, sandboxes from about eight cents an hour for a small VM, and training per million tokens. Prime Intellect names no SSO, audit log or security attestation.
Vendor details
Canonical URL
https://www.primeintellect.ai
Category
Agent infrastructure
Subcategory
Distributed training and RL compute platform
Funding status
Independent. Its homepage lists Founders Fund among its backers.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
An OpenAI-compatible inference API routing to multiple model providers, Hosted Evaluations against any OpenAI-compatible endpoint, the open source verifiers and prime-rl libraries, a REST API, CLI and Python SDK for sandboxes, GPU instances from more than fifty providers, Slurm for clusters, and browser environments through Browserbase.
In practice
You have an agent but no rigorous way to measure it beyond prompt tweaks. You package the task as a Prime Intellect environment with a scoring rubric and run Hosted Evaluations against your model and a few alternatives.
You want to train a small open model to use tools well on your task. You run Hosted Training on Lab with a LoRA recipe, pay per million tokens, and serve the adapter behind an OpenAI-compatible endpoint.
Your agent generates code you would not run on your own servers. You execute it in Prime Sandboxes, isolated microVMs with network allow lists and secrets that never reach logs.
Sources & related URLs
Research sources
Agentic Index coverage score
6.5 / 14 capabilities · 46%
| Integrations & Tool Calling | Partial |
|---|---|
|
Agents built in Prime Intellect environments reach outside systems through the tools their harness gives them. Environments declare tools, sandboxes and stateful tool environments. Hosted evaluations can be granted temporary sandbox, instance or tunnel permissions for environments that use tools, and Prime Tunnel exposes local services to them. Prime Intellect names no native connectors, OAuth actions or credential scoping for business systems, so tools are code the customer writes into the environment. Sourcedocs.primeintellect.ai/hosted-training/what-is-labread 2026-09-22 |
|
| Workflow Orchestration | Partial |
|
Lab runs agent steps in a loop the customer defines. An environment's harness sets the model's tools, sandboxes, context management and multi turn interaction. For each Hosted Training run, a dedicated Orchestrator manages environment logic, schedules rollout requests and coordinates training. Multi node clusters can be orchestrated with Slurm. The harness and rollout orchestration serve training and evaluation and are written in code. Prime Intellect names no workflow runtime that sequences, branches, retries and routes a deployed agent's production work. Sourcedocs.primeintellect.ai/hosted-training/what-is-labread 2026-09-22 |
|
| Knowledge Grounding & RAG | Not documented |
|
An environment's dataset is a set of tasks with optional answers for training and scoring. In the search agent guide, the customer builds its own document search inside an environment. Prime Intellect names no retrieval structure it maintains over the customer's content. Sourcedocs.primeintellect.ai/hosted-training/what-is-labread 2026-09-22 |
|
| Human Oversight & Guardrails | Not documented |
|
Prime Intellect names no approval step, consent checkpoint or escalation to a person for the agents that run in its environments or sandboxes. The developer configures network allow and deny lists per sandbox, which restrict where an agent's code can reach. Sourcedocs.primeintellect.ai/sandboxes/overviewread 2026-09-22 |
|
| Security, Identity & Governance | Partial |
|
Sandboxes are microVMs isolated at the hardware level, with network allow and deny lists for egress. Secrets are injected encrypted and never appear in logs or API responses. API keys carry scoped permissions, such as Inference, and team members hold Admin or Member roles. Prime Intellect names no SSO, SCIM or audit log and publishes no SOC 2 or ISO attestation. Sourcedocs.primeintellect.ai/sandboxes/overviewread 2026-09-22 |
|
| Observability & Auditability | Partial |
|
Each run leaves a record. Hosted evaluations store every run with its results in Prime Evals, stream logs to the CLI and dashboard, and can be shared. Hosted Training reports run metrics. Reserved clusters come with Grafana dashboards and alerting on GPU, node and network health. Prime Intellect names no audit log, step by step trace retention or SIEM export. Sourcedocs.primeintellect.ai/tutorials-environments/hosted-evaluationsread 2026-09-22 |
|
| Memory & State Persistence | Not documented |
|
What persists on Prime Intellect is compute state. Paused instances keep a workspace directory, and persistent disks hold datasets and checkpoints. Sandbox snapshots that save, restore and fork a sandbox mid run are marked coming soon and have not shipped. Prime Intellect names no memory an agent reads across runs. Sourcedocs.primeintellect.ai/sandboxes/overviewread 2026-09-22 |
|
| Deployment & Data Residency | Full |
|
Customers choose where their Prime Intellect compute runs. Before provisioning, GPU availability can be filtered by region (for example United States, Canada or Europe West), by data center and by provider cloud, across more than fifty providers. The training and evaluation stack (prime-rl and verifiers) is open source and runs on the customer's own hardware. Hosted Training and Hosted Evaluations run on infrastructure Prime manages, and no region is published for them. Sourcedocs.primeintellect.ai/api-reference/check-gpu-availabilityread 2026-09-22 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
The Environments Hub is a community registry of environments, ready to run, for training and evaluation. They span math, coding, games, search and tool use, and multimodal tasks, and the Hub also holds validated benchmark implementations. Lab guides give training recipes for code generation, tool use and search agents. Environments are tasks and scoring for research. The Hub carries no production workflows or agents built for a specific role. Sourcedocs.primeintellect.ai/tutorials-environments/environmentsread 2026-09-22 |
|
| Triggers & Channel Coverage | Not documented |
|
Work on Prime Intellect starts when a person or the customer's own code calls it. Hosted evaluations and training runs launch from the dashboard or CLI, and sandboxes and instances start through the API or SDK. Prime Intellect names no schedule, event subscription or other way to start an agent's work without a caller. Cluster alerting notifies people about hardware health, not agents. Sourcedocs.primeintellect.ai/tutorials-environments/hosted-evaluationsread 2026-09-22 |
|
| Model Flexibility & Routing | Full |
|
The customer chooses the model throughout. Prime Inference is an OpenAI-compatible API that routes requests to a range of model providers. Evaluations run against any model on it or any OpenAI-compatible endpoint the customer brings. Hosted Training lets the customer pick among open weights models (Qwen, Llama, NVIDIA Nemotron, gpt-oss and others), each with its own token price. Trained LoRA adapters deploy behind an OpenAI-compatible endpoint. Sourcedocs.primeintellect.ai/inference/overviewread 2026-09-22 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Outside code reaches Prime Intellect through a REST API (instances, availability, disks, clusters, sandboxes and chat completions) with scoped API keys, the prime CLI and a Python SDK for sandbox lifecycles. With the open source verifiers and prime-rl libraries, customers build and publish their own environments to the Environments Hub. The API, CLI and SDK are stable. Prime Intellect publishes no MCP server. Sourcedocs.primeintellect.ai/llms.txtread 2026-09-22 |
|
| Testing, Debugging & Optimization | Full |
|
Evaluating the customer's agent is a core product. An environment packages a dataset of tasks, the harness the model runs in (tools, sandboxes, multi turn interaction) and a rubric of reward functions and metrics that scores it. Hosted Evaluations run that environment against Prime Inference or any OpenAI-compatible endpoint, from the dashboard or CLI, and store each run with its logs and results in Prime Evals. The Environments Hub adds validated benchmarks such as AIME, MATH-500 and Humanity's Last Exam. Teams can test an agent on these datasets before production and score it the same way on every run. Sourcedocs.primeintellect.ai/tutorials-environments/hosted-evaluationsread 2026-09-22 |
|
| Browser & Computer Use | Not documented |
|
Browser use exists only inside training environments. The verifiers library's BrowserEnv lets an agent navigate and act on web pages, either in a DOM mode through Stagehand or in a computer use mode based on vision. The browser itself runs on Browserbase, a separate service with its own account and billing, sold beside the platform and not part of it. Prime Sandboxes are VMs for running code. Prime Intellect hosts no browser or desktop of its own for an agent to operate on web interfaces or desktop environments. Sourcedocs.primeintellect.ai/guides/browser-environmentsread 2026-09-22 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Prime Intellect rewrote Prime Agent, its open source coding agent, in Rust, adding native Windows support in beta and installation through Homebrew. Each session now runs in its own worker process so one crash leaves the others running, and sessions persist on disk so clients can reattach after a restart.
Bears on: Deployment / data residency
View sourcePrime Intellect publicly released Prime Inference, a serving platform for open models on its own GPU fleet, with serverless endpoints, reserved capacity and failover between data centers, starting with GLM-5.3. It enforces declared tool schemas during generation so tool calls stay valid through long agent sessions.
Bears on: Deployment / data residency
View sourcePrime Intellect made Prime Sandboxes generally available through its CLI and SDK and as part of its reinforcement learning stack. Each sandbox runs a Linux microVM with its own kernel, supporting Docker Compose, background services and custom Docker images. Accounts start with capacity for 1,024 concurrent sandboxes.
Bears on: Agent capability
View sourcePricing
Usage based: sandboxes from about $0.08 an hour (1 vCPU, 2 GiB, 10 GB), GPUs per hour by provider, training per million tokens
Prepaid credits drawn by usage: GPU instances per hour by provider, sandboxes per vCPU, GiB and GB hour, Hosted Training per million tokens by model, and inference per token
Included quota
No included allowance documented; usage draws on prepaid account or team credits.
What is public
Sandbox resource rates and Hosted Training token prices per model are published in the docs; GPU instance prices are quoted live by provider; reserved clusters are by request.
Billing mechanics
Usage draws down prepaid personal or team credits: GPU instances per hour (deducted every minute), sandboxes per resource hour, Hosted Training per million input, output and training tokens, and inference per token. Teams can share credits and set a billing email.
Cost watchouts
Instances bill per minute while active and are deleted automatically if credits run out, so auto top-up matters; sandboxes bill for CPU, memory and disk while running; training bills input, output and training tokens separately.
Variable cost rationale
Cost is driven almost entirely by usage, the GPU hours consumed, the providers chosen, inference volume, and the complexity and length of training and reinforcement learning runs, so spend scales directly with compute demand.
Additional watchouts
Set up auto top-up: instances are deleted when credits run out. Estimate training cost from token counts per model before a large run.
Overage / add-ons
Pay as you go from prepaid credits; there are no plans or overage tiers.
Sales call required
Mixed (some tiers require a call)
Free / trial
No free tier found on the pages read; the open source verifiers and prime-rl libraries are free to run on your own hardware
Commercial notes
Positioned as open infrastructure to democratize training, so much of the stack is open source and free to self host, while managed compute, inference, and training are usage based. Partners with clouds such as Nebius for elastic frontier hardware.
Key ambiguities
GPU prices vary by provider, card and region and are quoted live, so there is no single headline rate; total cost splits across instances, sandboxes, inference and Hosted Training.
Missing data
Inference per-model rates and reserved cluster pricing were not on the pages read; no pricing page exists at primeintellect.ai/pricing.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Prime Intellect
The closest documented capability profiles to Prime Intellect among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- LabWired6.5 / 14Adds documented Memory & State Persistence
- Guardrails AI7.0 / 14Adds documented Human Oversight & Guardrails
- Morph7.5 / 14Adds documented Knowledge Grounding & RAG
- AgentOps5.0 / 14Fuller documented coverage on Observability & Auditability
- Vercel Sandbox6.0 / 14Adds documented Memory & State Persistence
- Arcade7.5 / 14Adds documented Human Oversight & Guardrails
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded