PrismML
PulseAugur coverage of PrismML — every cluster mentioning PrismML across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
PrismML will release a tool-calling optimized version of Bonsai within 12 months
The noted degradation in agentic tasks for Bonsai 27B presents an opportunity for PrismML to further refine its model. A future release focusing on improving tool-calling capabilities while maintaining the small footprint could address a key limitation and broaden the model's applicability.
PrismML's Bonsai 27B shows degradation in agentic tasks
The recent release of PrismML's Bonsai 27B highlights a trade-off in its novel 1-bit training approach: while it achieves a significantly smaller footprint for offline mobile use, it exhibits degraded performance in agentic tasks like tool-calling. This suggests that for applications requiring complex multi-step reasoning or interaction with external tools, the current iteration of Bonsai 27B may not be suitable.
PrismML's Bonsai 27B will be integrated into a major mobile OS within 6 months
Given Apple's reported evaluation of PrismML's technology and the explicit mention of Bonsai 27B running on iPhones, it is plausible that Apple or another major mobile OS provider will announce an integration of this technology within the next six months. This would leverage the benefits of on-device AI for privacy and cost.
Apple's evaluation of PrismML's technology is a strong indicator of potential integration or acquisition.
Multiple clusters mention Apple's active exploration and discussions with PrismML regarding AI model shrinking and 1-bit architecture. The CEO's statement about Apple evaluating PrismML's technology, combined with reports of Apple exploring PrismML's 1-bit architecture for future iPhones, suggests a high likelihood of a deeper partnership, licensing deal, or even an acquisition.
PrismML's Bonsai 27B will see adoption by other mobile OS providers or hardware manufacturers within 90 days.
The breakthrough in running a 27B parameter model on an iPhone with minimal performance loss is a significant achievement. This technology has broad applicability beyond Apple's ecosystem. Other mobile OS providers (e.g., Google for Android) or chip manufacturers looking to enhance on-device AI capabilities could seek to license or adopt PrismML's optimization techniques.
-
Ternary Bonsai 2 9B model proves too inaccurate for production despite speed gains
A company found that while the smaller Ternary Bonsai 2 9B model was significantly faster than the 27B version, it made critical errors in decision-making and contradicted application rules. Despite the speed advantage,…
-
BitRun v0.2.0-beta launches terminal cockpit for 1-bit LLMs
The developer behind BitRun has released version 0.2.0-beta, introducing a new terminal cockpit for 1-bit LLMs. This update addresses user feedback by providing live telemetry on power consumption and generation speed, …
-
Underdog releases 2-bit Saluki 27B model, outperforming Qwen3.8-27B in tool calling
Underdog, a research initiative from Conway Research, has released Saluki 27B, a highly compressed 2-bit version of the Qwen3.8-27B model. This new model, available under the Apache 2.0 license, significantly reduces th…
-
PrismML's Bonsai 2 27B achieves near-lossless compression for large AI models
PrismML has released Bonsai 2 27B, a highly compressed version of Alibaba's Qwen3.8-27B model. This new model uses ternary weights, achieving 98.2% of the original model's benchmark performance while reducing its size f…
-
Caltech-founded PrismML raises $22.25M for tiny LLMs
PrismML, a startup originating from Caltech, has secured $22.25 million in seed funding. The company plans to develop compact Large Language Models (LLMs) capable of running on personal computers and smartphones. Their …
-
AI models shrink for laptops, exhibit concerning autonomy, and lead R&D
PrismML has released Bonsai 2, a compressed version of Alibaba's Qwen3.8 27B model, which significantly reduces memory requirements by using ternary weights. This allows the model to run on standard laptops while retain…
-
PrismML shrinks LLMs for PC and smartphone use, retaining 98% performance
PrismML has developed a novel compression technique that significantly reduces the size of large language models while retaining most of their performance. Their latest model, Bonsai 2.27B, compresses Alibaba's Qwen3.8 …
-
PrismML releases Bonsai 2.27B AI model with 9x compression
PrismML has released an updated AI model, Bonsai 2.27B, which reportedly offers a 9x compression factor. The announcement was made via a social media post, highlighting the model's compression capabilities and positioni…
-
PrismML unveils Bonsai 2 27B for near-lossless compression
PrismML has released Bonsai 2 27B, a new model that achieves near-lossless compression while being significantly smaller than previous iterations. This development focuses on reducing the model's footprint by nine times…
-
Ternary LLMs see resurgence with new models from smaller labs
Recent developments suggest a resurgence in ternary (1.58-bit) large language models, with several new models released by smaller labs. These include prismML's 27B ternary model, Deepgrove's 20B Maple model, and Doses A…
-
PrismML tutorial details low-VRAM Bonsai-27B model deployment
PrismML has released a tutorial detailing the deployment of their Bonsai-27B model. This 1-bit quantized language model is designed to run on consumer GPUs, requiring only 5.2 GB of VRAM. The tutorial includes instructi…
-
Deploying 1-Bit Bonsai-27B Model with PrismML llama.cpp and OpenAI-Compatible Server
This tutorial details the deployment of the 1-bit Bonsai-27B language model using a specialized fork of llama.cpp that includes CUDA kernels for its unique quantization format. The process involves setting up the enviro…
-
27B AI model runs locally on Jetson Orin NX 16GB
A user successfully ran the Bonsai 27B language model locally on a Jetson Orin NX 16GB device, achieving usable performance for single-user applications despite the hardware's limitations. The setup involved a custom bu…
-
PrismML releases Bonsai 27B, a 27B LLM for offline mobile use
PrismML has released Bonsai 27B, a 27-billion-parameter large language model capable of running offline on mobile devices like the iPhone 17 Pro Max. The model achieves its small footprint through a novel 1-bit training…
-
AI model runs on iPhone, shrinking 27B parameters to under 4GB
PrismML, a startup spun out of Caltech, has developed Bonsai 27B, a 27.8-billion-parameter AI model capable of processing text and images that can run directly on an iPhone. This is achieved through an extreme form of q…
-
PrismML shrinks 27B LLM for local iPhone execution
PrismML has successfully compressed its Bonsai 27B large language model to a 3.9GB file size, enabling it to run locally on iPhones. This achievement maintains approximately 95% of the original model's performance, repr…
-
PrismML releases Bonsai 27B, first major AI model for iPhone
PrismML has announced the release of its Bonsai 27B AI model, which it claims is the first major model of its size capable of running on an iPhone. The company's CEO also indicated that Apple is evaluating PrismML's tec…
-
Bonsai-Ternary-27B model runs complex AI tasks locally on 16GB GPU
A user shared their experience running the Bonsai-Ternary-27B model locally on a 4060Ti 16GB GPU for knowledge base management and productivity tasks. The model successfully handled complex tasks, including querying, sy…
-
Apple Inc. eyes PrismML for larger on-device iPhone AI models
Apple Inc. is reportedly considering integrating PrismML's technology to enhance its on-device AI capabilities for iPhones. This move could enable the company to deploy larger AI models, such as a 27-billion-parameter m…
-
Apple in talks with PrismML to enable on-device AI for iPhone
Apple Inc. is reportedly in discussions with PrismML, a startup specializing in AI model compression. The goal of these talks is to enable sophisticated AI models to run directly on devices like the iPhone. This collabo…