Alternatives
Products that do what Bonsai 1.7B ternary model at 442T/s on M4 Max does
We took a recently released Bonsai 1.7B ternary model from PrismML (https://github.com/PrismML-Eng/Bonsai-demo) and ran our agentic evolution search on it for 6 hours to optimize the Metal kernels. The search was fully autonomous. Measured against unmodified upstream llama.cpp at the same Bonsai/Q2_0 commit, same M4 Max: - tg128: 309.82 → 442.42 t/s (+42.0%) - pp512: 4250.32 → 4622.63 t/s (+8.8%)
- 11B
Mar 2026 · prismml.com
- 2BA
Introducing Bonsai 0.5B, one of the first ternary-weight LLMs to rival full-precision models of similar size, such as Qwen 2.5 0.5B and MobileLLM 0.5B. Trained on just 3.8B tokens, using 1,000x less data than other models, Bonsai redefines what’s possible for ultra-efficient training in low-bit models. Next, we're building larger and more powerful ternary-weight models for the edge. Technical Report: https://github.com/deepgrove-ai/Bonsai/blob/main/paper/Bonsa... Model (Unpacked): https://huggingface.co/deepgrove/Bonsai Reach us:…
2025 · github.com
- 3RP
The excitement surrounding PrismML’s 1-bit/ternary Bonsai models has the industry closely watching how smartphone giants, particularly Apple, will implement LLMs on edge devices. Moving AI on-device is a brilliant and necessary strategy. It ensures absolute user privacy in alignment with EU regulations, fundamentally shifts the economics away from costly cloud inference, and paves the way for a significant hardware upgrade supercycle as users seek true AI-capable silicon. To create a smart on-device "Semantic Router," models need to reach the 27B+ parameter scale. Achieving this on a…
Jul 2026
- 4MP
Aug 2026 · deepgrove.ai
- 5
- 6
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
27d ago · cactuscompute.com
- 7IB
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
Apr 2026 · github.com
- 8RA
Aug 2026 · github.com
- 9
- 10NW
Hey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok/s prefill and 1200 tok/s decode on consumer devices. We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention…
May 2026 · github.com
- 11
- 12
- 13WM
We wrote our inference engine on Rust, it is faster than llama cpp in all of the use cases. Your feedback is very welcomed. Written from scratch with idea that you can add support of any kernel and platform.
2025 · github.com
- 14
- 15
- 16
- 17CO
Hey HN, Henry and Roman here - we've been building a cross-platform framework for deploying LLMs, VLMs, Embedding Models and TTS models locally on smartphones. Ollama enables deploying LLMs models locally on laptops and edge severs, Cactus enables deploying on phones. Deploying directly on phones facilitates building AI apps and agents capable of phone use without breaking privacy, supports real-time inference with no latency, we have seen personalised RAG pipelines for users and more. Apple and Google actively went into local AI models recently with the launch of Apple Foundation Frameworks…
2025 · github.com
- 18
- 19
- 20TO
Mar 2026 · github.com
- 21
- 22
- 23DD
We recently used DeepSeek V4 Flash as a teacher for finance tasks with GPT-OSS-120B. Distillation works well on this problem. At a constrained 8k token budget, our self-distilled 120B scores 83.61% on FinanceReasoning, above Kimi K3 (81.93%) and Inkling (65.13%). We released the 20B open weights. With V4 as the teacher though, we realized it would be timely to measure if the censorship characteristic of it transferred to the distilled version of the base model. tl;dr it didn't, the teacher answered politically sensitive questions 7 SDs differently than expected, but the distilled model's…
Jul 2026 · ctgt.ai
- 24FT
Aug 2026 · github.com
Ranked by how close each launch is in meaning, then by votes. Refine with a description →