Up to 8 images; entire JSON body under 4.5 MB; up to 10,000 patches of 32 × 32 pixels across images; longest side at most 100× shortest.Source for Input limits (opens in a new tab)
Base64 data, state and questions all count toward body size. No numeric native text context ceiling verified.
Each question bills its text and all supplied images again. Image tokens = ceil(ceil(width/32) × ceil(height/32) × 1.5), per image, per question.Source for Billing scope (opens in a new tab)
A 1024×1024 image adds 1,536 input tokens per question. Two questions add 3,072 image tokens plus their text.
Hosted selector; no dated immutable snapshot is published in the reviewed docs.
Access
Liquid organization API key at POST https://api.liquid.ai/decisions/v1/systemone. Third-party gateways require separate contract review.
Regions
No model-specific processing-region contract verified in the reviewed sources.
INPUT / 1M TOKENS$0.042
OUTPUT / 1M TOKENS$0.00
Per billed input token; no generated output tokens. Each question is charged for its text and all supplied images again. The one-decision estimate uses billed text input; image conversion and multi-question totals are not calculated automatically. Promotional credits are not the ongoing tariff. Cached input price: not collected.
The hosted-service terms license customer inputs and outputs for model training, fine-tuning, evaluation and service improvement. The privacy policy covers hosted APIs and retention as necessary for its stated purposes, without a universal request TTL. Separate deployment agreements may differ. The pricing page’s custom-deployment no-training promise is not a blanket hosted-API policy.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
Liquid API
d1-ultra
Not verified
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
How it works
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
Minimum reusable input
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
Lifetime & refresh
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
How to enable it
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
Usage fields
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
Billing rules
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
Limits & data boundaries
Not collected
No native inference-cache specification established. Shared request state, parallel questions and data retention are not prompt caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 16 Oct 2026
What the measurements say
No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Managed hosted model selector under commercial API terms. Kept distinct from downloadable d1-3B and d1-omni-600M. The reviewed sources do not resolve the exact serving checkpoint or establish that its complete weights are independently private; this service entry is not counted as a proven private foundation-model checkpoint.
