Only 14MB! Cactus Project Enables LLMs on Phones and Wearables
cactus-compute · github · 2026-08-12
The open-source project needle (developed by cactus-compute) offers a foundation model solution specifically designed for tiny devices.
By aggressively compressing the model size to just 14MB, it enables local execution on hardware with limited compute and memory, such as smartphones, wearables, smart home appliances, and robots, pushing forward the accessibility of on-device AI.
Related event: Cactus Launches 14MB Needle 2 Edge AI Model(3 posts)→
More from Infra
- Nvidia and BlackRock's GPU Securitization Criticized as a $500B Irrational Mania — TiernanRayTech · 2026-08-12
- vLLM Compressor v0.13.0 Introduces MoE Expert Pruning and Arbitrary Bit-Width Quantization — vllm_project · 2026-08-12
- GB300 Rentals Hit $9/hr: Nebius Short-Term AI Capacity Deals Surge — zephyr_z9 · 2026-08-12
- Ignore the FUD: AI Compute Fundamentals Are Accelerating, Proven by Earnings — firstadopter · 2026-08-12
- Taiwan AI Server Exports Hit $9.7B in a Single Month, Surging 63% YoY — tengyanAI · 2026-08-12
- MiniCPM5 Trained Entirely on Huawei Ascend Tops Sub-2B Open-Weight Models — pstAsiatech · 2026-08-12