Bonsai ImagePrismML
|
Mercury 2Inception
|
|||||
Related Products
|
||||||
About
Bonsai Image Ternary 4B MLX 2-bit is a ternary-weight text-to-image diffusion transformer deployment for Apple Silicon. It is built as a quality-oriented Bonsai Image variant, using ternary {−1, 0, +1} transformer weights with FP16 group-wise scaling in the matrix-heavy transformer layers, including Q/K/V projections, output projections, and MLP weights. The model reduces the FLUX.2 Klein 4B transformer from 7.75 GB FP16 to a 1.21 GB Bonsai Image transformer, a 6.4× smaller footprint, while keeping visual quality and prompt fidelity close to the original model. The Apple Silicon deployment payload is 3.88 GB, including the MLX 2-bit diffusion transformer, a 4-bit Qwen3-4B text encoder, and an FP16 Flux2 VAE. After prompt encoding, the text encoder is offloaded, so the denoising loop only keeps the compact transformer and VAE resident. The model uses a 4-step FlowMatchEuler sampler with guidance 1.0 and shift 3.0, with no CFG and no negative prompts required.
|
About
Mercury 2 is the first reasoning model fast enough to pick up the phone, a reasoning diffusion language model built for real-time voice agents. Instead of making callers wait through seconds of dead air while an autoregressive model generates thinking tokens one by one, Mercury 2 uses a diffusion large language model architecture to generate tokens in parallel, decoding 1000+ tokens per second on standard NVIDIA GPUs. That speed is fast enough to run a full reasoning pass and start speaking within the latency budget of a natural conversation, reducing the cost of reasoning from seconds of silence to roughly 300 milliseconds. Mercury models work by corrupting clean text into noise, then training a standard Transformer to reverse the process and predict clean text across all positions simultaneously. Because each denoising pass touches many tokens, generation uses the GPU more efficiently than one-token-at-a-time decoding, making custom-silicon-like speed possible on NVIDIA H100s.
|
|||||
Platforms Supported
Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook
|
Platforms Supported
Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook
|
|||||
Audience
Apple Silicon developers building private local image-generation apps that need compact diffusion models
|
Audience
Voice AI infrastructure teams that need low-latency reasoning models for phone agents, tool-calling workflows, and natural customer conversations
|
|||||
Support
Phone Support
24/7 Live Support
Online
|
Support
Phone Support
24/7 Live Support
Online
|
|||||
API
Offers API
|
API
Offers API
|
|||||
Screenshots and Videos |
Screenshots and Videos |
|||||
Pricing
No information available.
Free Version
Free Trial
|
Pricing
No information available.
Free Version
Free Trial
|
|||||
Reviews/
|
Reviews/
|
|||||
Training
Documentation
Webinars
Live Online
In Person
|
Training
Documentation
Webinars
Live Online
In Person
|
|||||
Company InformationPrismML
Founded: 2026
United States
prismml.com
|
Company InformationInception
United States
www.inceptionlabs.ai/blog/mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone
|
|||||
Alternatives |
Alternatives |
|||||
|
|
|
|||||
|
|
|
|||||
|
|
|
|||||
|
|
|
|||||
Categories |
Categories |
|||||
Integrations
Cerebras
GPT-4.1
Groq
Inception Labs
LiveKit
OpenAI
Pipecat
Retell AI
Vapi AI
|
Integrations
Cerebras
GPT-4.1
Groq
Inception Labs
LiveKit
OpenAI
Pipecat
Retell AI
Vapi AI
|
|||||
|
|
|