Darkroom Vision
VISION · 30B
Reads images and answers without filters. Multimodal, sealed in hardware.
Open a room$0.22 in · $0.95 out /Mtok
Context
64K tokens
Architecture
30B · ~3B active · MoE · vision
Modality
text + vision
Speed
balanced
Privacy
Sealed · confidential
Filter
Uncensored
Input
$0.22 / Mtok
Output
$0.95 / Mtok
About
A 30B vision-language model (~3B active, MoE) that sees. Drop in screenshots, photos, diagrams, or documents and ask anything — visual reasoning, OCR, and image understanding with no content filter. 64K context, fully sealed.
Best for
Image understandingOCRVisual reasoningDocumentsimage input
Sealed in hardware
In production, every sealed session runs inside an attested TDX enclave — your prompt decrypts in the room, the answer comes back, and each session carries its own measurement you can verify against an open DCAP verifier.
room fingerprint published at launch · tdx