Prime Intellect
GPU marketplace and RL training stack behind a monospace, terminal-styled site
Tier A · StrongReferenceServiceprimeintellect.ai

Prime Intellect sells an integrated stack for training and serving custom models: a Lab for post-training agents with RL environments and evaluations, serverless inference with LoRA support, and on-demand or reserved GPU capacity brokered across 50+ providers from 1 to 256 GPUs. Its research arm publishes open-source work including Prime Agent, INTELLECT-3, and SYNTHETIC-2, and the CLI installs with `pip install prime`. It earns a place in this catalog for its site language rather than its infrastructure: monospace type, numbered navigation labels (TRAINING01, INFERENCE02, COMPUTE03, RESEARCH04), a live careers count baked into the nav, and posted hourly GPU rates. It is the cleanest current example of the terminal-brutalist aesthetic that AI-infrastructure brands have converged on.
pip install primeFor an agent
Use this as a visual reference when a brief calls for a technical, research-lab register — monospace body copy, numbered section labels, dense specification tables, near-zero decoration — and specifically when you need to see that style applied to a real pricing and product page rather than a portfolio piece. It is narrower than Modal as a reference: one marketing site, no docs-site design system and no published tokens, so do not lean on it for component patterns. If the task is compute rather than design, docs.primeintellect.ai serves llms.txt and the org's open repos (prime-agent, verifiers) are the substantive artefacts, not the landing page.
Why it is here. It is the canonical instance of the terminal-brutalist AI-lab visual language, useful precisely because that style is now widely copied and rarely executed with this much restraint.
Facts
- License
- MIT, Apache 2.0, Proprietary / all rights reserved
- Pricing
- Usage-based / credits, Paid, Free but account required
- Framework
- Python
- Styling
- Unknown
- Distribution
- Hosted SaaS product, Marketplace listing, CLI installer, Public HTTP API, GitHub repository
- Platform
- Web
- Content type
- Code, Prose / writing, Dataset / structured records
- Agent readiness
- Structured documentation, Public API without auth, CLI installable, Readable open source
- Accessibility
- Unknown
- Maturity
- Actively maintained
- Pricing detail
- Posted on-demand hourly GPU rates: H200 $1.99–$3.14/hr, B300 $4.99/hr, B200 $3.49/hr. Reserved three-year cluster pricing is quoted separately via 'Book a call'.
- GitHub stars
- 287
- Package
- prime
Helps you build
6Notable for
- numbered navigation labels (TRAINING01, INFERENCE02, COMPUTE03)
- monospace-first typography across a commercial site
- hourly GPU rates published on the homepage
- live open-role count in the nav
- open-source research repos: Prime Agent, INTELLECT-3, SYNTHETIC-2
Provenance
Verified 2026-08-15. HTTP 200. https://www.primeintellect.ai (200, nav labels, product sections, and GPU rates read), https://www.primeintellect.ai/llms.txt (404), https://docs.primeintellect.ai (200, redirects to /introduction), https://docs.primeintellect.ai/llms.txt (200), GitHub API for PrimeIntellect-ai/prime (287 stars, MIT, pushed 2026-08-15) and the org repo list
Also in Design Systems & Brand Guidelines
6- ARIA Authoring Practices Guide · W3C's 30 canonical widget patterns with keyboard maps and working HTML examples
- Branding Style Guides · Archive of ~3,840 real brand guideline PDFs, filterable by year, region and industry
- Material Design 3 · Google's open design system: 39 components, tokens, and Compose/Web code
- Modal · Python serverless GPU platform whose site is a dev-tool marketing design reference
- NameThatUI · Visual dictionary that maps 76 web and macOS UI elements to their real names and code
- The Component Gallery · 60 UI components cross-referenced against 95 real design systems