Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.
Context / input limit
Not collected
No model-specific public value established in the reviewed sources.
Maximum output
Not collected
No model-specific public value established in the reviewed sources.
Context caching
Not collected
No model-specific public value established in the reviewed sources.
Reasoning
Not collected
No model-specific public value established in the reviewed sources.
Tool calling
Not collected
No model-specific public value established in the reviewed sources.
Structured output
Not collected
No model-specific public value established in the reviewed sources.
Streaming
Not collected
No model-specific public value established in the reviewed sources.
Batch processing
Not collected
No model-specific public value established in the reviewed sources.
Fine-tuning
Not collected
No model-specific public value established in the reviewed sources.
Parameters
Not collected
No model-specific public value established in the reviewed sources.
Architecture
Diffusion Transformer; identity reference tokens concatenated with visual tokens. Production model distilled to two diffusion steps.Source for Architecture (opens in a new tab)
Knowledge cutoff
Not collected
No model-specific public value established in the reviewed sources.
Languages
Not collected
No model-specific public value established in the reviewed sources.
Release date
Not collected
No model-specific public value established in the reviewed sources.
Endpoint pricing has not been collected. Check the provider for applicable rates.
Data & enterprise terms
Hosted on AWS. Customer data deletion is described as taking up to 90 days, with residual backups subject to their lifecycle. Service-improvement terms permit usage analysis within specified customer-instruction and consent restrictions; this is not a universal no-training or zero-retention promise.
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.
Synthesia
Not verified
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
How it works
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
Minimum reusable input
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
Lifetime & refresh
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
How to enable it
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
Usage fields
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
Billing rules
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
Limits & data boundaries
Not collected
Saved avatars, reusable videos and enrollment reuse are not documented prompt caching or discounted input-token billing.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
Caching reviewed 14 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
Synthesia identifies this named model in its proprietary research or product materials. No public model weights or permissive weight license are established; open client code does not expose the serving model.
