PrismML
An AI startup incubated at the California Institute of Technology, focused on large language model compression technology and backed by Khosla Ventures. Its core technology enables the compression of large models to a size suitable for deployment on mobile devices.
Timeline (last 90 days)
PrismML released Ternary Bonsai 2 9B, whose GGUF file is only 2.25 GiB, while the 27B version is 6.70 GiB.
BitRun v0.2 can run PrismML's 8-billion-parameter model bonsai-8b entirely in CPU memory, using about 2.08 GB of memory, without needing a discrete GPU, a cloud API, or compilation.
PrismML's Ternary Bonsai 2 applies ternary quantization to the Qwen3.8 27B model, has received 3.87 million downloads, compresses the language model from 54GB to 5.9GB — a 9.3x reduction — and officially claims it retains 98.2% of the FP16 model's capability, with an average score of 84.78 on 14 reasoning benchmarks.
PrismML announced the deployment of its self-developed micro large language model onto smart glasses powered by Qualcomm chips.
PrismML's core philosophy is to build open-weight AI that can run on-device.
The specific model scale, performance metrics, and mass production timeline of PrismML's smart glasses project have not yet been officially disclosed.
Bonsai 2, released by PrismML, is a heavily quantized version based on Qwen 3.8 27B.
The test was run on a machine equipped with 1x RTX 5090 32GB, using PrismML's llama.cpp fork.
The poster is one of the founders of PrismML and the related toolchain (atomic.chat fork), and the evaluation carries a potential conflict of interest.
PrismML's specific technical parameters, benchmark results, and micro model scale have not yet been disclosed in currently available public materials.
1 more timeline events
All Facts (8)
BitRun v0.2 can run PrismML's 8-billion-parameter model bonsai-8b entirely in CPU memory, using about 2.08 GB of memory, without needing a discrete GPU, a cloud API, or compilation.
50%UnverifiedPrismML's Ternary Bonsai 2 applies ternary quantization to the Qwen3.8 27B model, has received 3.87 million downloads, compresses the language model from 54GB to 5.9GB — a 9.3x reduction — and officially claims it retains 98.2% of the FP16 model's capability, with an average score of 84.78 on 14 reasoning benchmarks.
50%UnverifiedPrismML's core philosophy is to build open-weight AI that can run on-device.
50%UnverifiedThe poster is one of the founders of PrismML and the related toolchain (atomic.chat fork), and the evaluation carries a potential conflict of interest.
50%UnverifiedPrismML is an AI lab attempting to change the industry's use of AI with an extremely small large language model.
50%UnverifiedPrismML released Ternary Bonsai 2 9B, whose GGUF file is only 2.25 GiB, while the 27B version is 6.70 GiB.
50%UnverifiedPrismML announced the deployment of its self-developed micro large language model onto smart glasses powered by Qualcomm chips.
50%UnverifiedBonsai 2, released by PrismML, is a heavily quantized version based on Qwen 3.8 27B.
50%