WH

Whittle-Reflex-25B-A3B

WhittleGGUF by logic65Updated 2026-10-10
Download on Hugging Face
Parameters25.6B
Context256K
Layers40
Architectureqwen4exp
Smallest quant15.7 GB
Quick answer

Whittle-Reflex-25B-A3B needs about 16.8 GB of memory at Q4_K_M (a 15.7 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 3 quants, from 15.7 GB (Q4_K_M) to 27.2 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 2.7 GB.

Whittle-Reflex-25B-A3B – a 25.6B parameter System 1 decision model

Whittle-Reflex-25B-A3B is a 25.6 billion parameter mixture‑of‑experts model designed for System 1 decision tasks. It is not a chat model and only returns calibrated probabilities for a set of options. The model is available in GGUF format for llama.cpp.

Good at

  • Option selection
  • Reasoning tasks
  • Tool usage

Limits

  • No chat generation
  • Requires 8 k context for memory estimates
  • Early preview – may change

Best for

  • Decision‑making pipelines
  • Evaluation of multiple-choice datasets
  • Automated reasoning workflows

Who it’s for. Researchers and developers needing a fast, non‑generative decision model. The model is released under Apache‑2.0 and can be run with official llama.cpp.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_K_M15.7 GB 16.8 GB
Too big Get file →
Q6_K21.1 GB 22.3 GB
Too big Get file →
Q8_027.2 GB 28.4 GB
Too big Get file →

For 8 GB VRAM use Q4_K_M (file size 15.67 GB, estimated 16.8 GB memory). For 12 GB VRAM use Q6_K (21.09 GB, 22.3 GB). For 16 GB VRAM use Q8_0 (27.21 GB, 28.4 GB). Estimates are approximate.

What GPU do you need for Whittle-Reflex-25B-A3B?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is Q6_K (21.1 GB file, about 22.3 GB in use).
  • 32 GB VRAM: the best quant that fits is Q8_0 (27.2 GB file, about 28.4 GB in use).

Which Whittle-Reflex-25B-A3B quant should you choose?

Q4_K_M is the usual starting point for Whittle-Reflex-25B-A3B: it balances file size and quality well. Here it is a 15.7 GB file.

The Q8_0 file is 27.2 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architectureqwen4exp
Parameters25.6B (approximate, mixture of experts)
Active parameters3B
Layers40
Trained context262,144 tokens
KV cache (fp16)80 KB per token

Source

Quantized bylogic65
FamilyWhittle
Licenseapache-2.0
Downloads (HF)651
Last updated2026-10-10

Frequently asked questions about Whittle-Reflex-25B-A3B

How much VRAM does Whittle-Reflex-25B-A3B need?

About 16.8 GB at Q4_K_M with an 8K context window. That is the 15.7 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run Whittle-Reflex-25B-A3B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 16.8 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Whittle-Reflex-25B-A3B GGUF?

It depends on the quant: Q4_K_M is 15.7 GB, Q6_K is 21.1 GB, Q8_0 is 27.2 GB.

What context length does Whittle-Reflex-25B-A3B support?

The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.7 GB at 32K.

Where can I download Whittle-Reflex-25B-A3B GGUF?

The files are on Hugging Face in the logic65/Whittle-Reflex-25B-A3B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does Whittle-Reflex-25B-A3B use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run Whittle-Reflex-25B-A3B on 8 GB VRAM?

Yes, the Q4_K_M quantized file requires about 16.8 GB of VRAM at 8 k context, which is within 8 GB if using a smaller context or swapping.

Is this model suitable for chat?

No, it is a decision model only and does not generate chat responses.

What license applies to this model?

The model is released under the Apache‑2.0 license.

Similar models

Browse all →

More from logic65

All by logic65 →

More Whittle models

All Whittle →

Other medium models

All medium models →