Token pricing
Per 1M tokens, by the base model Facet builds on. Every workspace's first 5 million tokens each month are free, on any plan. Usage is billed monthly.
| Base model | Input | Output |
|---|---|---|
| Qwen3.5 0.8B | $0.04 | $0.12 |
| Qwen3.5 4B (default) | $0.08 | $0.25 |
Training
- Drafting a model is free.
- Every build is quoted before it starts. The quote is a cap: if a build would pass it, it stops and asks.
- Most first versions cost $1 to $3. Your first one trains free when its quote is $15 or less.
- A later fix to the instructions alone costs cents; one that needs a new build costs about what the first version did.
- Stopping a build charges only what ran. Exporting a model is free.
Next to frontier models
What a month of calls costs at list token prices, for a call of 1,000 tokens in and 200 out. Facet's row uses the default base and takes off the free tokens. Building the model is quoted separately.
| A month of | 10,000 calls | 100,000 calls | 1,000,000 calls |
|---|---|---|---|
| Facet, Qwen3.5 4B | $0.76 | $12 | $129 |
| Claude Opus 5.5 | $80 | $800 | $8,000 |
| Claude Sonnet 5.5 | $40 | $400 | $4,000 |
| Claude Haiku 5.5 | $2.00 | $20 | $200 |
| GPT-5.5 | $110 | $1,100 | $11,000 |
| GPT-5.6 Sol | $80 | $800 | $8,000 |
| GPT-5.6 Terra | $44 | $440 | $4,400 |
| GPT-5.6 Luna | $4.40 | $44 | $440 |
| GPT-4o | $45 | $450 | $4,500 |
| GPT-4.1 mini | $7.20 | $72 | $720 |
| Grok 4.7 | $32 | $320 | $3,200 |
Token prices only. Frontier prices are OpenRouter list prices, checked November 12, 2026. A small model is worth it only if it does your task well, so Facet scores every version on your evals and can score it next to the frontier model you use today.
Questions
Is Facet free to try?
Yes. Drafting a model is always free, your first model trains free when its quote is $15 or less, with one retry if it fails, and every workspace gets 5 million tokens of calls a month on any plan. You need a card only for later builds and for calls past the free tokens.
What does it cost to build a model with Facet?
Every build is quoted before it starts, and the quote is the most it can cost. Most first versions cost $1 to $3. A later fix to the instructions alone costs cents; one that needs a new build costs about what the first version did. Stopping a build charges only what ran.
What does it cost to call a model built with Facet?
Calls are priced per million tokens by base model: Qwen3.5 0.8B is $0.04 in and $0.12 out, and Qwen3.5 4B is $0.08 in and $0.25 out. Usage is billed monthly to your card, and a month under $0.50 rolls into the next.
Can a Facet bill surprise me?
No. Every build shows its price before it starts, and that price is a cap it never goes past. A monthly spending limit, on by default, stops new work and calls when it is reached, and an Owner can raise it.
Is there pricing for high volume?
Yes. Past about 1 billion tokens a month, or for a dedicated GPU, talk to us. Write to support@halfspacelabs.com.
Does the price table mean a small model is as good as a frontier model?
No. The table compares token prices only. Whether a small model is good enough for your task is a question Facet answers with your evals, and it can score a version next to the frontier model you use today on the same hidden examples, before you switch.
The billing docs cover invoices, the spending limit and the free training credit.