On 14 July, Hacker News Best surfaced Bonsai 27B as the first 27B-class model to run on a phone. LocalLLaMA repeated the claim that day and described a 1-bit model with 8B parameters and an MMLU-R score of 65.7. The gap between the 27B-class billing and the stated parameter count remains unexplained. The story currently rests on two community signals from a single day.
If factual, the impact is 0%. Developers working on on-device AI gain another reported example of a compact language model running on phone hardware. Phone users have no stated product change or rollout.
Watch for original model documentation, reproducible phone specifications, independent benchmark results, and clarification of the 27B-class and 8B-parameter descriptions.
<!-- SC_OFF --><div class="md"><p>This is nuts.</p> <p><a href="https://huggingface.co/prism-ml/Bonsai-8B-gguf">prism-ml/Bonsai-8B-gguf · Hugging Face</a></p> <p>has anyone tested
<table> <tr><td> <a href="https://www.reddit.com/r/OpenAI/comments/1t4ffmo/openai_expected_to_produce_as_many_as_30_million/"> <img alt="OpenAI expected to produce as many as 30
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1s41egi/rfdetr_nano_and_yolo26_doing_ondevice_object/"> <img alt="RF-DETR Nano and YOLO26 doing on-device
<!-- SC_OFF --><div class="md"><p>Are local 4B models usable on smartphone?</p> <p>Just did a vibe check on a Pixel Pro 10, Gemma 4B vs Qwen 3.5 4B, starting from handheld photos
<!-- SC_OFF --><div class="md"><p>After burning way too much money on GPT-4 API calls, I finally went all-in on local models with Ollama. The progress in 2026 is insane
<table> <tr><td> <a href="https://www.reddit.com/r/LocalLLaMA/comments/1sskmhv/recent_open_models_from_last_6_months_nov_2025/"> <img alt="Recent Open models from last 6 Months -
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →