Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Decart / Oasis

Oasis 3 Preview

A hosted driving world model that generates camera frames from a scene prompt and throttle/steering actions.

World modelsVideo generation
previewClosed weightsWorld simulation
Official documentation (opens in a new tab)Reviewed 15 Oct 2026
9 of 23 specification fields collected
Output resolution768×512See specification notes below
Clip durationNot collectedAwaiting source review
Input billingSee ratesDecart Oasis API
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Maximum input tokens
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Maximum output
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Reasoning
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Tool calling
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Structured output
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Streaming
SupportedSource for Streaming (opens in a new tab)

Stateful gRPC action-to-frame session, distinct from Lucy WebRTC editing.

Batch processing
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Fine-tuning
Not collected

The RL example trains a driving policy using the world model; it does not establish customer fine-tuning of Oasis weights.

Knowledge cutoff
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Release date
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Parameters
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Architecture
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Languages
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Context caching
Not collected

Not established for this exact model and API. Media generation does not imply language-model token limits or capabilities.

Input modalities
Text, ActionsSource for Input modalities (opens in a new tab)
Output modalities
VideoSource for Output modalities (opens in a new tab)
Output resolutions
768×512Source for Output resolutions (opens in a new tab)
Release window
2026-06Source for Release window (opens in a new tab)

Month-level launch announcement; exact date not established.

Output formats
VP9, RGB uint8 framesSource for Output formats (opens in a new tab)
Original generation context
A new scene prompt resets world context and the action rollout.Source for Original generation context (opens in a new tab)
Motion controls
Each infer call takes four [throttle, steering] actions in [−1, 1] and returns four frames per stream.Source for Motion controls (opens in a new tab)
Output structure
Three streams: left_forward, front and right_forward.Source for Output structure (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

Decart Oasis API

Provider source (opens in a new tab)
API model ID
oasis-3-preview
Access
Hosted preview with API-key access; no session started. Exact entitlement and regional availability not independently tested.
Regions
Not established
API endpoint URL
https://oasis-grpc.decart.aiSource for API endpoint URL (opens in a new tab)

gRPC over HTTP/2; TLS by default. Not the Lucy REST or WebRTC route.

Authentication
Decart API key supplied to A2VClient or DECART_API_KEY environment variable.Source for Authentication (opens in a new tab)
SDK support
decart-oasis Python SDK; Python 3.10 or later.Source for SDK support (opens in a new tab)
Processing workflow
Initialize → Prompt → Infer → Finish. Session routing uses x-session-target after initialization.Source for Processing workflow (opens in a new tab)
Processing timeout
Default per-RPC timeout: 120 seconds. Not a session-duration or inference-latency guarantee.Source for Processing timeout (opens in a new tab)
Billing basis
USD 0.02 per active generation second; enterprise pricing may differ.Source for Billing basis (opens in a new tab)

The generic price-table heading says 720p, but the specific model contract defines 768×512 frames.

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 15 Oct 2026
CategoryUsageRateUnit
Media generationWorld simulation1$0.02active generation second

Rate conditions

  1. Oasis 3 Preview; 768×512 camera frames. No additional per-stream multiplier is established. Enterprise arrangements may differ.
Pricing source (opens in a new tab)

Data & enterprise terms

Reviewed API terms permit input/output use for service improvement and training, and do not treat them as confidential. Enterprise-specific overrides were not reviewed.

Read the service policy (opens in a new tab)

Caching

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.

Decart Oasis API

oasis-3-preview
Not verified

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

How it works
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

Minimum reusable input
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

Lifetime & refresh
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

How to enable it
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

Usage fields
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

Billing rules
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

Limits & data boundaries
Not collected

No model prompt-cache contract or cache discount established. Upload reuse, client-token expiry and stateful video context are separate mechanisms.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 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

The provider names this hosted model, retains rights in its models and limits access through its API. No matching checkpoint appears in the reviewed official public-weight inventory. Closed-weight classification is an inference from these sources; it is not applied to Lucy Edit Dev releases or public SDK code.

Classification evidence (opens in a new tab)

Sources reviewed