Thinking Machines Inkling
Model Information
Thinking Machines Lab's open-weights multimodal reasoning model hosted on Together AI. Inkling natively processes text, images, and audio and can transcribe speech, follow spoken instructions, answer questions about recordings, and reason over longer-form audio. It has 975B total parameters with 41B active parameters.
Model ID
togetherai.thinkingmachines-inkling
Use this ID when making API calls to reference this model
Provider
togetherai
Model Type
multimodal
Accuracy Tier
premium
Release Date
July 15, 2026
Supported Languages
No language information available
Performance & Cost
Cost
$0.61200/hour
$0.00017/second
Maximum Duration
2m
Maximum File Size
5.25 MB
Features
Supported capabilities and functionalities
Core Features
Punctuation
Diarization
Streaming
Speaker Labels
Word Timestamps
Confidence Scores
Custom Vocabulary
Profanity Filtering
Noise Reduction
Voice Activity Detection
Subtitle Formats
SRT Support
VTT Support
Technical Specifications
Input/output formats and technical details
Subtitle Format Support
No subtitle formats supported
Supported Audio Encodings
WAV
Supported Sample Rates
16000 Hz