Overview
Each tier is called AgentykLM [Animal] v[Version], for example AgentykLM Jaguar v1. The small tiers answer in an instant and cost little, so they suit routing, short chats and high-volume extraction. The large tiers take on code, maths and multi-step reasoning, with context windows up to 256K tokens for whole contracts, dossiers and codebases. Nine tiers are available today through the Agentyk API with one key, and you can try them in your browser in AgentykCloud. Orca, the largest, is on the way.
The lineup
| Tier | Name | Size | Primary use |
|---|---|---|---|
| 01 | Hummingbird | Nano · Edge | On-device, mobile, instant text completion |
| 02 | Mongoose | Micro | Classification, intent detection, routing |
| 03 | Gazelle | Mini | Conversation, chatbots, low latency |
| 04 | Greyhound | Small | High-volume API, data extraction |
| 05 | Cheetah | Medium | Real-time translation, live support |
| 06 | Leopard | Medium-Large | Long-context document processing |
| 07 | Jaguar | Large | Complex reasoning, coding, math |
| 08 | Tiger | X-Large | Enterprise reasoning, multi-step logic |
| 09 | Grizzly | XX-Large | Multi-agent, scientific workloads |
| 10 | Orca | Ultra · coming | Frontier: the most complex problems |
Highlights
- Nine tiers available today, with Orca on the way
- Context windows up to 256K tokens on the long-document tier
- Express versions of Hummingbird and Jaguar for quick replies
- Streaming, tool calls and strict JSON output when a reply has to follow a fixed format
- Text embeddings and reranking for search over your own documents
- Works with the chat API your tools already use: change the address, add your Agentyk key, pick a tier
- One key for every tier, hosted in the EU