The frontier, included.
Everything else, connected.
Two kinds of models live in the same menu. Included models run on our accounts and are covered by your plan. Connected models are yours, from an account you hold to a model on your own hardware, and we never meter them.
Included with every plan
| Model | Maker | Good at | Context |
|---|---|---|---|
| Opus 4.8 | Anthropic | The heavyweight. Deep reasoning, long documents, careful judgment. | 1000k |
| Sonnet 5 | Anthropic | Fast frontier quality for writing and everyday thinking. | 1000k |
| Sonnet 4.6 | Anthropic | A steady all-rounder, kept for continuity. | 1000k |
| Haiku 4.5 | Anthropic | Answers in a blink. Where quick lookups go. | 200k |
| GPT-5.6 Sol | OpenAI | OpenAI's frontier flagship. Deep reasoning and code. | 1050k |
| GPT-5.6 Terra | OpenAI | The balanced GPT. Strong tools, strong everyday judgment. | 1050k |
| GPT-5.4 | OpenAI | The previous GPT generation, kept for continuity. | 400k |
| GPT-5.4 mini | OpenAI | Quick, capable, inexpensive. Carries routine work. | 400k |
| GPT-5.4 nano | OpenAI | The smallest GPT. Instant answers at the lowest cost. | 400k |
| Gemini 3.5 Flash | Google's newest. Fast, sharp, and huge context. | 1000k | |
| Grok 4.5 | xAI | xAI's flagship. A different temperament at full strength. | 500k |
| Gemini 3.1 Pro | Enormous context. Reads whole contracts and codebases. | 1000k | |
| Grok 4.3 | xAI | Long context and low cost, with Grok's directness. | 256k |
| Gemini 3.1 Flash Lite | Google's fast tier for high-volume simple work. | 1000k |
The catalog grows as new models ship. Every included model works with the router, Compare and Council.
Bring your own model
On your machine
ollama · lm studio · vllm
Run a model locally and point Heuricity at it. It joins the same menu with a hollow marker, chats with the same tools, and costs you nothing but electricity.
On your accounts
your key · mistral · deepseek · openai
Connect accounts you already hold. Their models run at your rates, under your agreements, and never touch your Heuricity allowance.
Through OpenRouter
one key · hundreds of models
One OpenRouter connection brings its whole catalog along. Heuricity fetches the model list for you.
For the technically curious
A connection is any OpenAI-compatible endpoint or a provider account key. Keys are encrypted at rest and never appear in logs or the browser. Connected models are first-class citizens: the router may pick them, Council may seat them, and every message shows its source, so a turn that ran on your connection says so right on the message.
Where a model runs is a visible property, never a guess. Included models run on our provider accounts under our agreements. Connection turns run on yours, are never metered by us, and are never used to subsidise anyone else. If a connection fails mid-conversation, Heuricity says so and offers the plan as a fallback rather than failing silently.