LocalClaw/LLM/Agents-A1
Open-weight MoE

Agents-A1

InternScience Apache 2.0 agentic VLM. 35B-A3B MoE, 262K context, strong long-horizon search/tool-use benchmarks and official Q4_K_M GGUF artifacts for local workstations.

32 GB power user 32 GB RAM Q4_K_M Local research agent
Run with LocalClaw Compare all models
Parameters
35B (3B active, MoE)
Minimum RAM
32 GB
Model size
21 GB
Quantization
Q4_K_M

Can Agents-A1 run locally?

Agents-A1 belongs on 32 GB machines when you want stronger quality without jumping to server hardware.

Search for agents-a1 in LM Studio or another GGUF-compatible runtime.

Model sourceInternScience/Agents-A1-Q4_K_M-GGUF
chatcodevisionagentreasoningpowertool-callingmultimodal

Install path

01
Check RAM fitMinimum 32 GB RAM. Start with the Q4_K_M quant.
02
Load the modelSearch agents-a1 in LM Studio.
03
Control locallyUse LocalClaw to manage models, agents, chat, channels and scheduled OpenClaw work.

Strengths

  • Official InternScience Apache 2.0 open-weight release
  • Strong fit for long-horizon search, scientific reasoning, engineering tasks and tool use
  • High Hugging Face activity across the base model and official GGUF variants
  • Official Q4_K_M GGUF package includes the multimodal projector for local llama.cpp-style runtimes
  • 262K context window for multi-step agent traces and long research workflows
  • Text-only serving mode can skip the vision encoder when memory is tight

Limitations

  • Workstation-class model: the Q4_K_M GGUF package is about 21GB before KV cache and vision/projector overhead
  • Very long context and multimodal prompts can require much more than the base Q4 footprint
  • Agent benchmark claims should be validated on the exact local runtime and tool schema
  • Not as laptop-friendly as 4B-14B general chat models despite sparse active compute

Best use cases

  • Local research agent
  • Tool-calling and API workflow rehearsal
  • Long-horizon search experiments
  • Scientific and engineering reasoning
  • Image-aware agent workflows
  • Workstation LM Studio and llama.cpp testing

Capability profile

speed
5
quality
9
coding
8
reasoning
9

Technical notes

Developer
InternScience
License
Apache 2.0
Context window
262,144 tokens
Architecture
Qwen3.5 MoE-based multimodal agent model with about 35B total parameters, a 35B-A3B sparse shape, vision/projector support and tool-use chat template support.

This model fits these next steps

Hardware fit is based on LocalClaw's RAM tier, model size and quantization metadata. Always leave memory headroom for your OS and runtime.

Comfortable headroomMac mini M4 Pro 48GBMobile workstationMacBook Pro M4 Max 36GBPower-user picks32GB RAM guide

Similar models to compare

Qwen AgentWorld 35B-A3B 35B (3B active, MoE)North Mini Code 1.0 30B (3B active, MoE)Qwen 3.6 35B-A3B 35B (3B active, MoE)OLMo 3 32B Think 32BDiffusionGemma 26B-A4B Instruct 26B (4B active, diffusion MoE)

Where to go next