Open
No minimum
−30%off the reference list price
Start with pay-as-you-go access. No minimum commitment.
DeepSeek V3.2 output: $0.28 per million
InferenceLeading models.Clear prices.
Open models from Kimi, GLM and DeepSeek through one familiar API. Pay for the tokens you use, with automatic discounts as your usage grows.
Prepaid usage. No minimum commitment. OpenAI-compatible.
One model.
More than one GPU.
Model shards run across a connected GPU group. Formal Engines coordinates the execution; your application receives one response.
One model
Split the model across a connected GPU group.
01020304
Formal Engines
Orchestration layer
Model memory · connectivity · priceGPU 01Model shard 1
GPU 02Model shard 2
GPU 03Model shard 3
GPU 04Model shard 4
Coordinated model execution
One response
Each GPU holds part of the model. The group works together to serve the request.
Illustrative layouts. GPU grouping depends on workload and connectivity.
Great models.
Clear prices.
USD per million tokens, against reference list rates. Every line is 30% off before volume tiers.
| Model | Context | Reference in / out | Formal Engines in / out | You save per 1M |
|---|---|---|---|---|
| DeepSeek V4 FlashDeepSeek | 1M | $0.06 / $0.12 | $0.04 | |
| DeepSeek V4 ProDeepSeek | 1M | $0.73 / $1.45 | $0.39 | |
| GLM 5.2Z.ai | 1M | $0.67 / $2.12 | $0.45 | |
| GLM 5.3Z.ai | 1M | $0.98 / $3.08 | $0.65 | |
| Kimi K3Moonshot AI | 1M | $2.10 / $10.50 | $1.80 |
Reference list rates in this card are dated 3 September 2026. The saving column assumes three input tokens for every output token. Use the estimator below to compare model and workload combinations.
Explore the full model catalogueHow the price is possible
A shared engine.
A simpler serving layer.
Formal Engines brings model access behind one endpoint. Your application chooses the model; the serving layer handles the request.
The wider platform coordinates distributed GPU capacity within price and placement policies. Explore the foundation behind inference and post-training.
Explore the compute engineFig. 2 / Where the capacity isGPU supply by price
More usage, more off.
Monthly thresholds use spend at Open-tier rates. Discounts are totals off the reference list price.
Scale
From $2,000 / month
−40%off the reference list price
A lower rate as your monthly usage grows.
DeepSeek V3.2 output: $0.24 per million
Volume
From $10,000 / month
−45%off the reference list price
For teams with sustained production traffic.
DeepSeek V3.2 output: $0.22 per million
Reserved
From $20,000 / month
−50%off the reference list price
Our deepest published discount for larger workloads.
DeepSeek V3.2 output: $0.20 per million
Estimate the bill.
Estimate your monthly spend for the discounted models shown.
Estimated monthly spendOpen tier
Reference list rates
$0
Formal Engines
$0
$0
saved per month · 0% below reference list rates
Tiers apply automatically as volume grows.
One base URL to change.
Python
from openai import OpenAI
client = OpenAI(
base_url="https://formalengines.com/inference/v1",
api_key="fe_live_your_key",
)
completion = client.chat.completions.create(
model="kimi-k3",
messages=[{"role": "user", "content": "Summarise this incident report."}],
stream=True,
)Set your Formal Engines API key and base URL, then choose a model from the rate card.
curl
curl https://formalengines.com/inference/v1/chat/completions \
-H "Authorization: Bearer fe_live_your_key" \
-H "content-type: application/json" \
-d '{"model":"deepseek-v4-pro","messages":[{"role":"user","content":"ping"}]}'Send chat completions through the same endpoint. See the documentation for supported request options.
Send us your current inference bill.
We will quote the same models at the same volume and latency target, then show you what the difference works out to before you move any traffic.