Tycho 1.0
When the task is clear, the volume is high, and every token matters.
High-Efficiency Scale
Classification and routing
Ticket triage, intent detection, document labeling, workflow decisioning at scale
Extraction and summarization
Structured fields from emails, PDFs, contracts, call transcripts, and support logs
Support automation
Draft responses, escalate edge cases, reduce repetitive queue volume
Significant savings with prompt caching (up to 90%) on repeated prompts.
Comparable to Anthropic Haiku 4.5 and GPT-5.4 Mini
Primary use cases
Route tickets, label documents, triage inbound requests, and identify intent at scale.
Pull structured fields from emails, PDFs, contracts, support logs, and operational records.
Summarize calls, cases, messages, documents, and workflow events.
Decide which workflow, tool, team, or model should handle a request.
Draft responses, classify intent, escalate edge cases, and reduce repetitive queue volume.
Execute simple tool-based workflows with predictable structure.
Retrieve and answer over internal knowledge where request volume is high.
Key capabilities
Lower cost for high-volume AI
Tycho is designed for the workloads where token volume becomes material: support, document processing, internal automation, enrichment, and product features used every day by thousands or millions of users.
Fast enough for production interfaces
Use Tycho where users expect responsiveness: chat flows, support tools, workflow assistants, classification systems, and inline application features.
Simple to deploy
Tycho is exposed through the same Radium API layer as Clarke and Hal, making it easy to route workloads by cost, latency, or complexity without changing application architecture.
Enterprise-ready by default
Tycho runs behind Radium's enterprise delivery layer, with controls for authentication, logging, monitoring, routing, governance, and deployment isolation.
For teams using Haiku or GPT mini-class models.
Tycho is designed for teams that need a fast, economical model for scaled production usage. If your current workload depends on Claude Haiku, OpenAI mini/nano-class models, or earlier small-model endpoints, Tycho is the Radium model to evaluate first.
Replace high-volume Claude and OpenAI workloads with Tycho.
Start with your highest-volume API calls. Move them to Tycho. Measure cost, latency, and quality. Scale from there.