24 GB accelerator memory
Q4_K GGUF · reduced context · llama.cpp
Weights and projector use about 20.5 GiB before runtime state, so 24 GB requires shorter context and single-session serving.
Huihui AI
Text model
Agents-A1 Abliterated Q4_K
A long-horizon agent model with fewer refusal behaviors in a local Q4_K build.
ChatCodingFewer refusals
Capabilities
What you can use it for.
The most useful reasons to choose this model, without making you read through its repository first.
Agentic loopsTool callingVisionFewer refusals
01
Long-horizon research
02
Tool-using agents
03
Multi-step engineering
For providers
Want to serve this model?
The provider requirements live here, separate from the information you need to choose and use the model.
Starting point24 GB GPU memory
Conservative estimate
48 GB accelerator memory
Q4_K GGUF + BF16 projector · llama.cpp
The exact Q4_K file is 19.7 GiB and the vision projector adds 0.84 GiB; 48 GB leaves practical room for context and provider concurrency.
Source and license
Check the original model.
OpenMayhem links back to the source repository so you can inspect the model card, files, license, limitations, and creator guidance directly.
Abliterated edition of InternScience Agents-A1