Releases·Model Release
Yandex releases AliceAI-Foundation-80B-A3B-Base, a custom-architecture Russian base model competing with Qwen and DeepSeek
Yandex has published AliceAI-Foundation-80B-A3B-Base on Hugging Face, a large MoE-style base model positioned as a Russian-built competitor to Qwen 3.5 and DeepSeek V4 Flash. Community posters note it uses a fully custom architecture rather than being a Qwen finetune. It ships as a base model without post-training, limiting immediate usability for chat applications.
415 engagement1 sourceMon, Sep 21, 2026, 07:24 PM
AI Overview
- Yandex's AliceAI-Foundation-80B-A3B-Base uses its own fully custom architecture, not a Qwen3 finetune, per its Hugging Face page
- The model ships as a base checkpoint only — Yandex has not post-trained it, unlike Qwen3.5/3.6
- No llama.cpp support exists for the model yet
Related
Social·Community
Reddit user seeks <40B local alternatives to banned Qwen/Deepseek models for coding and 120k-token document QA
A user asks for sub-40B open model alternatives to Qwen and DeepSeek, which are banned at their workplace, for coding and long-document QA tasks like questioning a 120k-token medical history. They suggest candidates including Gemma 31B, Muse Glimmer 30B, and Nemotron, and ask whether fine-tunes of these can beat Qwen3.8 27B at coding.
A user asks for sub-40B open model alternatives to Qwen and DeepSeek, which are banned at their workplace, for coding and long-document QA tasks like questioning a 120k-token medical history. They suggest candidates including Gemma 31B, Muse Glimmer 30B, and Nemotron, and ask whether fine-tunes of these can beat Qwen3.8 27B at coding.
44 engagement1 sourceredditFri, Oct 2, 2026, 08:44 AM
Releases·Tool Release
BigMoeOnEdge engine runs 92 GB DeepSeek V4 Flash on 12 GB RAM Android phone at 1 tok/s
A developer's engine, BigMoeOnEdge, successfully ran DeepSeek V4 Flash IQ2_M (92 GB) on a mid-range Android phone with 12 GB RAM at 1 token/s, demonstrating that large MoE models can run on edge devices. The engine supports many models including Gemma 26B and Qwen 3, and is built on llama.cpp for modularity.
A developer's engine, BigMoeOnEdge, successfully ran DeepSeek V4 Flash IQ2_M (92 GB) on a mid-range Android phone with 12 GB RAM at 1 token/s, demonstrating that large MoE models can run on edge devices. The engine supports many models including Gemma 26B and Qwen 3, and is built on llama.cpp for modularity.
26 engagement1 sourceredditTue, Aug 4, 2026, 07:01 PM
Social·Community
DeepSeek V4 Flash (UD-Q3_K_M) on a single RTX 4090 at 64k context — config and measured numbers
A user documents running DeepSeek V4 Flash in UD-Q3_K_M quantization on a single RTX 4090 with 64k context, providing exact hardware and software configuration. The setup requires 128 GB of RAM as a minimum, with the model file at 121 GB. This serves as a reference for others attempting similar configurations.
A user documents running DeepSeek V4 Flash in UD-Q3_K_M quantization on a single RTX 4090 with 64k context, providing exact hardware and software configuration. The setup requires 128 GB of RAM as a minimum, with the model file at 121 GB. This serves as a reference for others attempting similar configurations.
21 engagement1 sourceredditTue, Aug 4, 2026, 08:09 PM
Social·Communityfeatured
Users report DeepSeek-V4-Flash-0731 speeds on varied GPUs, 11–59.5 tok/s
Reddit users are sharing benchmark results for running DeepSeek-V4-Flash-0731 GGUF locally on consumer and workstation GPUs, with speeds ranging from 11 tok/s on a mixed AMD setup to 59.5 tok/s on an RTX PRO 6000 Max-Q eGPU. Quantization, multi-GPU setups, and DSpark drafter support are common themes, with some users noting improved performance from newer llama.cpp builds.
Reddit users are sharing benchmark results for running DeepSeek-V4-Flash-0731 GGUF locally on consumer and workstation GPUs, with speeds ranging from 11 tok/s on a mixed AMD setup to 59.5 tok/s on an RTX PRO 6000 Max-Q eGPU. Quantization, multi-GPU setups, and DSpark drafter support are common themes, with some users noting improved performance from newer llama.cpp builds.
Updated Wed, Aug 5, 2026, 12:53 PM — New posts show DSpark drafter support and vLLM on SM89 GPUs, with speeds up to 59.5 tok/s on RTX PRO 6000 Max-Q eGPU.
81112.2k engagement22 sourcesredditSun, Aug 2, 2026, 12:00 AM
Social·Community
Alibaba open-weights Qwen3.8, a frontier model that beats Claude and GPT on some jobs
A Product Notes blog post argues Qwen3.8 is not simply the best model but something better, characterizing Alibaba's open-weight frontier model as beating Claude and GPT on some jobs while losing on others. The author describes how they are actually using the model in practice.
A Product Notes blog post argues Qwen3.8 is not simply the best model but something better, characterizing Alibaba's open-weight frontier model as beating Claude and GPT on some jobs while losing on others. The author describes how they are actually using the model in practice.
1.5k engagement2 sourcesrss, xThu, Sep 3, 2026, 07:28 AM