Skip to content
UsecaseBoard
Saved0
← All use cases
huggingfaceopen-model

Tencent Hy-Memory-30B-A3B: open long-term memory writer for agents

At a glance

Tencent Hunyuan open-weights Hy-Memory-30B-A3B (~3B active MoE) to replace the general LLM on the write path of Mem0/MemOS/Hy-Memory—trained with on-policy distillation + attributed memory RL; beats Claude Opus 5 / DeepSeek V4.1 Flash on a fixed cross-harness memory-writer matrix; Apache-friendly HF weights + GitHub framework. This workflow is listed under open-model, agent-memory, MoE, Mem0, post-training.

Agents / tools
Hy-Memory-30B-A3B, Mem0, MemOS, Hy-Memory, vLLM

Building a tool for these workflows? Sponsor / inquire.

Save ideas in this browser. Upvotes count across the board.

Prompt excerpt

Serve hy-memory with vLLM as --served-model-name hy-memory, then swap it in as the Mem0/MemOS writer and re-run LoCoMo/LongMemEval samples against your prior writer.

Related workflows

Explore all →
huggingfaceopen-model

inclusionAI Ling-3.1-flash: 560B MIT hybrid MoE for agents and coding

inclusionAI/Ant Ling open-sourced Ling-3.1-flash (560B total / ~25B active, 1M context, MIT) for agents, office, search, and SWE—day-0 SGLang/vLLM recipes, tool/reasoning parsers, and strong Terminal-Bench / SWE-bench Pro / SkillsBench numbers; works in Claude Code, Hermes, OpenClaw, and related harnesses.

Ling-3.1-flashSGLangvLLM+2
Source ↗
productopen-model

Paper Instruments Ultramarine 1 Flash: open model for knowledge work

Paper Instruments launched Ultramarine 1 Flash for professional knowledge work (finance, law, consulting, ops)—API at $0.15/$0.50 per 1M in/out, claims Fable-class JobBench 62.9 and ~60x cheaper than Fable per task; delivers Excel/PPT/Word and ERP/MES workflows; weights coming soon on Hugging Face, training code MIT.

Ultramarine 1 FlashPaper Instrumentsdocument agents
Source ↗
blogopen-weights

Reflection Beam 1.0: 501B open-weight agentic coding model

US lab Reflection open-sourced Beam (501B MoE / 23B active, 1M context, Apache 2.0) for coding and long-horizon agents, with day-0 vLLM/SGLang/Transformers support, Mirror CLI harness, and 40+ inference partners including OpenRouter and Ollama.

Beam 1.0Mirror CLIvLLM+3
Source ↗
blogdecision-models

IBM Granite Switch 4.1: chat, decisions, and ASR on one endpoint

IBM Granite Switch 4.1 (Oct 15, HF blog): 8B/30B compose chat + Decision aLoRA + Guardian safety aLoRA + Granite-Speech-5.0-Turbo-CTC ASR in ~base VRAM via Activated LoRAs with shared KV cache—swap chat/decision/speech on one endpoint; vLLM plugin and llama.cpp support; community can mix Granite Libraries adapters.

Granite Switch 4.1IBM GraniteGranite Speech+2
Source ↗