What Radium does

Radium (legal entity: Radium Enterprise AI) is an enterprise AI inference provider incorporated in Ontario, Canada, with an operational base in San Francisco. The company offers a swap-in, OpenAI- and Anthropic-compatible inference platform targeting enterprise teams that are already integrated with frontier-model APIs and are seeking to reduce AI infrastructure spend. Radium's go-to-market positions the service as a cost- and performance-optimized alternative for production AI workloads, claiming approximately 50% lower cost than comparable OpenAI and Anthropic endpoints at equivalent capability.

The platform is anchored by a proprietary bare-metal infrastructure stack designed from the kernel level up for AI workloads. Core technical components include a zero-virtualization kernel optimized for GPU locality and predictable performance (marketed as ~0 jitter), an RL-based intelligent orchestration engine that performs workload-aware scheduling across NUMA, SXM, PCIe, and fabric topologies, and a custom dual-plane network fabric separating InfiniBand AI traffic from Ethernet control/storage. Radium delivers three model tiers — Hal 1.0 (maximum capability, comparable to Opus 4.7 and GPT-5.5), Clarke 1.0 (balanced performance, comparable to Sonnet 4.6 and GPT-5.4), and Tycho 1.0 (high-efficiency scale, comparable to Haiku 4.5 and GPT-5.4 Mini) — all built on open-source foundations and exposed through an OpenAI-compatible API contract that allows single-endpoint migration.

Radium generates revenue through usage-based, per-token API pricing across its three model tiers, with input and output tokens billed separately. Distribution combines a self-serve signup platform (deploy.radium.cloud) with direct enterprise sales and an OEM/embedded channel demonstrated through the Realbotix partnership. Named reference customers include Square (R&D prototyping), EQTY Lab (climate model training, COP28), Realbotix (humanoid robotics inference), and Alexi (legal-tech retrieval models), spanning financial services, climate, robotics, and legal verticals. Pricing examples indicate approximately $1,600/month for 320M tokens on the highest tier versus $3,200/month for equivalent OpenAI or Anthropic endpoints, with optional prompt caching providing up to 90% savings on repeated prompts.

Name

Radium

Legal name

Radium Enterprise AI

Website

https://radium.cloud

Company type

Private

Founded year

2020

Operating status

Operating

Headcount range

11–50 employees

Short description

Radium provides enterprise AI inference through a bare-metal infrastructure platform, offering OpenAI-compatible API access to three model tiers (Hal, Clarke, Tycho) at approximately half the cost of frontier providers, targeting enterprise teams seeking to reduce AI infrastructure spend.

Ownership category

akta.pro rank

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Infrastructure, Technology or R&D, Personnel, Operations, Marketing or Sales

Revenue model

1. API Token-based Inference: Radium generates revenue through consumption-based API access to AI model inference endpoints. Customers pay per token (input and output tokens separately) based on which model tier they use (Hal, Clarke, or Tycho). The pricing is structured to be approximately 50% lower than comparable OpenAI and Anthropic endpoints. Additional cost optimization comes from prompt caching which can save up to 90% on repeated prompts.

Pricing tiers

Go-to-market motion1 record

Distribution channels3 records

Marketing channels5 records

Core offering

Radium is an OpenAI/Anthropic-compatible AI inference platform that delivers three open-source-based model tiers (Hal 1.0, Clarke 1.0, Tycho 1.0) over a serverless API. Customers access the models by swapping the endpoint URL—a single line of code—and pay per input and output token at approximately 50% of comparable frontier provider pricing, with optional prompt caching that reduces costs on repeated prompts by up to 90%.

Product overview

Radium is an enterprise AI inference platform offering three distinct model tiers built on open-source foundations and delivered through proprietary bare-metal infrastructure. The core portfolio includes: (1) Hal 1.0 for maximum capability complex reasoning and agentic workflows, (2) Clarke 1.0 for balanced production workloads, and (3) Tycho 1.0 for high-efficiency, high-volume tasks. All models are delivered via a unified serverless inference platform with OpenAI/Claude-compatible API, supported by Radium's zero-virtualization bare-metal infrastructure that eliminates abstraction layers between software and hardware for optimized GPU utilization and reduced latency.

Differentiator

Problem solved

Functional benefit

Brands

• Hal 1.0: Maximum capability model tier for complex reasoning, agent workflows, and software tasks.
• Clarke 1.0
• Tycho 1.0

Products and services

• Hal 1.0 Hal 1.0 is Radium's maximum capability AI model tier for complex reasoning, agentic workflows, code generation, architecture-level planning, and multi-file software tasks. It is comparable to Anthropic Opus 4.7 and OpenAI GPT-5.5 and is priced at $2.25 per million input tokens and $11.50 per million output tokens on a pay-as-you-go basis, accessed via Radium's OpenAI/Claude-compatible API.
• Clarke 1.0 Clarke 1.0 is Radium's balanced performance AI model tier for general-purpose enterprise workloads, including document reasoning, structured outputs (JSON), code generation and debugging, RAG, and enterprise copilot applications. Comparable to Anthropic Sonnet 4.6 and OpenAI GPT-5.4, priced at $1.50 per million input tokens and $7.00 per million output tokens on pay-as-you-go.
• Tycho 1.0 Tycho 1.0 is Radium's high-efficiency AI model tier for high-volume, cost-sensitive workloads such as classification, extraction, summarization, ticket triage, intent detection, document labeling, and lightweight agent execution. Comparable to Anthropic Haiku 4.5 and OpenAI GPT-5.4 Mini, priced at $0.50 per million input tokens and $2.25 per million output tokens on pay-as-you-go.

Quantifiable outcome

• 2x lower inference latency vs AWS (Carnegie Mellon University benchmark)
• +4 more outcomes

Named customers4 records

Segments3 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

AI capability9 records

Feature4 records

Partnerships

Four partnerships are on record, tiered core.

• RealbotixcoreTechnology or Integration · 7 August 2025Realbotix Corp. (TSX-V: XBOT) announced a strategic collaboration with Radium to power real-time AI companions using Radium's serverless inference platform, addressing critical latency challenges in consumer robotics. The partnership enables lightning-fast conversations and GPU auto-scaling for Realbotix's humanoid robots built for entertainment, customer service, and personal well-being applications.
• Stanford UniversitycoreStrategic or Co-development PartnerStanford University's Center for Research on Foundation Models conducted independent benchmarks comparing Radium to GCP. The research validated that Radium achieved 30% higher Model FLOP Utilization (50-55% vs 39-45% for GCP) on identical JAX workloads.
• Massachusetts Institute of Technology (MIT)coreStrategic or Co-development PartnerMIT researchers contributed to research suggesting that open models achieve roughly 80-90% of the performance of leading closed models at a fraction of the cost, which supports Radium's value proposition.
• Carnegie Mellon UniversitycoreStrategic or Co-development PartnerCarnegie Mellon University conducted independent benchmarks comparing Radium to AWS. Research validated 2x lower inference latency vs AWS and 58% better utilization at 16-layer depth vs AWS, with the performance advantage compounding as models get deeper.

Scale indicators4 records

Recent moves6 records

Expansion highlights5 records

Direct peers

• Fireworks AI: Fireworks AI provides serverless AI inference for open-source and fine-tuned models with OpenAI-compatible APIs, targeting enterprise developers seeking lower latency and cost. Directly comparable to Radium in pricing model, customer segment, and inference-optimization focus.
• OctoAI: OctoAI provides optimized AI inference infrastructure for running open and custom models at scale, with a focus on cost and latency reduction. Closely aligned with Radium's bare-metal optimization thesis and enterprise inference positioning.
• DeepInfra: DeepInfra offers low-cost, low-latency inference APIs for open-source LLMs and other models, targeting enterprise and developer use cases. Direct overlap with Radium on cost-optimized inference and OpenAI-compatible APIs.
• Together AI: Together AI operates an open-source-focused AI inference and training cloud with API endpoints positioned as lower-cost alternatives to OpenAI/Anthropic. It is the closest direct competitor to Radium in product, target customer (enterprise AI teams), and go-to-market motion (API-first, OpenAI-compatible).
• Modal: Modal provides serverless cloud infrastructure for AI/ML workloads with developer-focused APIs and auto-scaling. Comparable to Radium's serverless inference offering and emphasis on developer ergonomics for production AI deployments.
• Lambda: Lambda operates GPU cloud infrastructure purpose-built for AI training and inference, including reserved and on-demand GPU clusters. Comparable to Radium on the bare-metal AI infrastructure layer, though Lambda is more vertically integrated in hardware.
• Replicate: Replicate runs a cloud for running and deploying open-source machine learning models via API, with usage-based pricing similar to Radium's token-based model. Targets developers and enterprises seeking frictionless access to AI models without managing infrastructure.
• Anyscale: Anyscale offers AI compute and inference infrastructure built on Ray, serving enterprise teams running production AI workloads with a focus on performance and cost efficiency. Overlaps with Radium's enterprise AI infrastructure positioning and developer-first motion.

Broad incumbents

• CoreWeave: CoreWeave is a large-scale GPU cloud provider offering AI training and inference infrastructure to enterprise customers. It is a broader incumbent in the same category as Radium's underlying infrastructure, though CoreWeave's portfolio extends well beyond inference-specific workloads.
• AWS Bedrock: AWS Bedrock is a managed service providing access to multiple foundation models via API on AWS infrastructure. It is the broad incumbent competitor against which Radium benchmarks its 2x latency and 58% utilization advantages.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks5 records

Key highlights6 records

Customer concentration

• Xx.com/radium_ai
• LinkedInlinkedin.com/company/radium-llm

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds4 records

Investors4 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Radium