Frontier-class models, without the frontier bill.
The same class of models you build on today, running on infrastructure we own and operate, for about half of what the closed labs charge.

Radium models are designed for workloads you run everyday
Code
Autocomplete, refactors, tests, and review, running on the same class of model your team already uses.
Support
Customer conversations answered in your voice, on your policies, at the volume support actually runs.
Retrieval
Answers pulled from your own documents, with the source attached and the context window to carry it.
Copilots
The tools your team works in all day, answering questions about your own work, for everyone on it.
Hal 1.0
Comparable to Anthropic
Opus 5.5 or OpenAI GPT-6 Astra
Clarke 1.0
Comparable to Anthropic
Sonnet 5 or OpenAI GPT-6 Sol
Tycho 1.0
Comparable to Anthropic Haiku
4.5 or OpenAI GPT-6 Luna
The same workloads, at a lower cost to operate
Hal 1.0
Opus 5.5
GPT-6 Astra
Radium's most capable model for complex reasoning, multi-step agents, and deep code analysis.
Anthropic's most capable model for complex reasoning, coding, and agentic workflows.
OpenAI's flagship reasoning model for advanced agents and coding tasks.
$11.50 / output MTok
$20 / output MTok
$50 / output MTok
Clarke 1.0
Sonnet 5
GPT-6 Sol
Radium's default production model for RAG, copilots, agents, and customer-facing AI.
Anthropic's balanced model for production reasoning and coding at scale.
OpenAI's multimodal flagship model for general-purpose and coding tasks.
$7.00 / output MTok
$10 / output MTok
$10 / output MTok
Tycho 1.0
Haiku 4.5
GPT-6 Luna
Radium's most cost-efficient model for classification, extraction, summarization, and support automation.
Anthropic's fastest model, built for low-latency, high-volume tasks.
OpenAI's small, low-cost model for lightweight, high-volume tasks.
$2.25 / output MTok
$5 / output MTok
$0.50 / output MTok
One line of code to switch.
A different class of performance.
Every new account gets $25 in credit. No card required.

Measured against the models it can replace.
One integration speaks both the OpenAI and Anthropic dialects and serves all three models. The two formats are drifting apart, and moving between them is turning into real work. Radium answers in whichever one your stack already speaks.
Read our benchmark reportYour workload runs on hardware we own.
We own and operate
every layer
Your models run on infrastructure that is ours end to end, not rented with a markup on top.
Your data trains nothing
What you send stays yours, and is not used to train anyone's next model.
Regional hosting
Choose where your workloads run, for teams that answer to data-residency requirements.



