> ## Documentation Index > Fetch the complete documentation index at: https://docs.edge0.ai/llms.txt > Use this file to discover all available pages before exploring further. # Models: 8B vs 35B Edge0 offers two preview tiers. For a first try, or when you want to reduce the download size and resource footprint, start with 8B. When your device has enough resources, try 35B and compare the two on your typical tasks. > **Edge0 8B** > > `Edge0-8B-A1B-preview` · First try; resource-limited devices > **Edge0 35B** > > `Edge0-35B-A3B-preview` · Experiencing a larger model when resources allow ## Specification comparison | Item | Edge0 8B | Edge0 35B | | - | - | - | | Full name | `Edge0-8B-A1B-preview` | `Edge0-35B-A3B-preview` | | Parameter scale | About 7.9B total, about 1.2B active | About 35B total, only a small subset active per step | | Base model | `inclusionAI/Ling-3.0-tiny-base` | `Qwen/Qwen3.6-35B-A3B` | | Layers | 24 | 40 | | Routed experts | 128 (plus 1 always-resident shared expert) | 256 (plus 1 always-resident shared expert) | | Experts used per step | 8 | 4 | | Release format | 4-bit, with LoRA and prerouter files | 4-bit, with LoRA and prerouter files | | Best for | First try; resource-limited devices | Experiencing a larger model when resources allow | B means billion. Parameter count, model file size, and runtime memory are different measures: an MoE model uses only part of its parameters per step, and quantization reduces file size, so the 35B model does not need 35 GB of memory. Both tiers are currently `preview` releases. The model cards note that tool use, multi-step planning, and long-horizon autonomous tasks are not yet well optimized. ## Model size The model file format and size differ by platform: | Platform or file | 8B | 35B | | - | -: | -: | | Released 4-bit model files | About 4.2 GB | About 23 GB | | Model in the Windows app | About 5 GB | About 21 GB | | iOS app including the model | About 5 GB | About 20 GB | | Android converted GGUF files | Depends on the actual conversion | About 21.7 GB | The macOS **Models** page shows the model size. For storage requirements, see [Device Requirements](/device-requirements). ## Speed reference The table below shows generation speed on each platform's test device, in tokens/s. The test devices and conditions differ by platform and are not meant for direct side-by-side comparison; actual speed is also affected by input length, memory use, and device temperature. | Platform and test device | 8B | 35B | Notes | | - | -: | -: | - | | macOS: MacBook Air M3 | 18–20 | 10–12 | | | Windows: i7-14700K, RX 9070 GRE, 48 GB DDR5, Windows 11 | \~44.8 | \~27.7 | Measured with the model fully loaded in memory; a PC with less memory will be slower | | iOS: iPhone 16 Pro, iOS 26.6.2 | 10.9 | 6.4 | | | Android: Lenovo TB322FC, Snapdragon 8 Elite, 16 GB RAM | 29–32, declining to about 10 within 480 s | 6–9 | Affected by device heat during long runs | Model formats used in the Android test: 8B is a Q8-class GGUF + LoRA, and 35B is a mixed int8 MoE (demand-paged). The wait before a reply begins is different from the generation speed. The wait is longer for the first load or for longer inputs. ## Choosing by device Your device Requires: Open the guide → * **macOS**: both tiers run on a MacBook Air M3, with resident memory of about 1.7 GB and 4 GB respectively. * **Windows**: 8B needs 8 GB or more of memory; 35B needs 16 GB or more. * **iOS**: you can bundle one tier or both, and only the selected model is loaded at a time. * **Android**: 8B needs 8 GB or more of memory; 35B needs 12–16 GB. For a first try, prepare and import 8B first. For full platform requirements, see [Device Requirements](/device-requirements). ## Official download links | Model | Hugging Face | ModelScope | | - | - | - | | Edge0 8B | [Edge0-8B-A1B-preview](https://huggingface.co/Edge0/Edge0-8B-A1B-preview) | [Edge0-8B-A1B-preview](https://www.modelscope.cn/models/Edge0/Edge0-8B-A1B-preview) | | Edge0 35B | [Edge0-35B-A3B-preview](https://huggingface.co/Edge0/Edge0-35B-A3B-preview) | [Edge0-35B-A3B-preview](https://www.modelscope.cn/models/Edge0/Edge0-35B-A3B-preview) | After downloading, prepare the model files following the platform instructions in [Download & First Run](/download-and-first-run). For license details, see [License & Model Terms](/license-and-model-terms). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.