Models & Pricing
Nomic runs on frontier foundation models from Anthropic and Google, plus Nomic's own domain-specific models for parsing and embedding engineering documents. This section explains how usage is billed, how far a typical AI Usage allocation goes, and where to look for per-model and indexing details.
How AI usage is billed
Nomic contracts include an AI Usage pool, denominated in USD, that is shared across your organization and refills on your billing cycle. Every AI operation deducts from this pool at the underlying provider's rate:
- Assistant & Workflows — Billed per million tokens at each foundation model's API rate. Input, output, cache read, and cache write tokens are metered separately. The cost of a turn depends on the model, how much context Assistant pulls in, and how long the response is.
- Indexing (parsing + embeddings) — Billed per page parsed by Nomic's domain-specific Parse model. See Indexing.
Every transaction is recorded as a SpendEvent and surfaced in Admin → Usage. See Analytics & monitoring for the dashboard.
How much does $20 in AI usage buy you?
The AI Usage pool is sized in dollars because different models have very different prices. The numbers below assume the recommended default — Claude Sonnet 4.6 — and are approximate; actual cost depends on context size, number of tool calls, and response length. Heavy caching pushes these higher; very long uncached research queries push them lower.
| Workflow | Est. cost per run | Runs per $20 |
|---|---|---|
| Quick Q&A on an indexed project (single tool call, short answer) | ~$0.02 – $0.05 | 400 – 1,000 |
| Typical Assistant thread (5–10 messages, ~10 tool calls, citations) | ~$0.10 – $0.25 | 80 – 200 |
| Deep Research across a large project (extended tool use, long synthesis) | ~$0.50 – $1.50 | 13 – 40 |
| Drawing review on a single sheet | ~$0.15 – $0.40 | 50 – 130 |
| Code compliance check on a ~50-page spec section | ~$0.30 – $0.80 | 25 – 65 |
| Submittal review against a project spec (multi-file, structured output) | ~$0.75 – $2.00 | 10 – 25 |
Rules of thumb for stretching your $20
- Default to Sonnet 4.6. Best cost/quality tradeoff for most queries. Reach for Opus only on the hardest reasoning, and Haiku for quick exploration.
- Keep projects scoped. Smaller projects and tags mean less input context per turn.
- Reuse threads. Cache reads are roughly 10× cheaper than fresh input tokens, so continuing a session is dramatically cheaper than starting a new one.
- Use Deep Research selectively — it runs many more tool calls and produces longer answers.
- Index once, query many. After a file is indexed, every question only pays for the foundation-model tokens it actually uses. See Indexing.
Where to next
- Models — The foundation models in the Assistant picker and what they cost per 1M tokens.
- Indexing — How Nomic Parse + Embed work, per-page pricing, and how to manage indexing spend.
- Admin → Usage — In-product dashboard for auditing AI spend in real time.