CA

Caelum G4 38B A12.5B

CaelumGGUF by mradermacherUpdated 2026-08-09
Download on Hugging Face
Parameters38B
Context256K
Layers48
Architecturegemma4
Smallest quant0.0 GB
Quick answer

Caelum G4 38B A12.5B needs about 36.5 GB of memory at Q4_K_M (a 23.2 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 24 quants, from 0.0 GB to 31.2 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 51.5 GB.

Caelum G4 38B A12.5B – a 38B Mixture‑of‑Experts Gemma 4 model

Caelum G4 38B A12.5B is a 38.01B parameter Gemma 4 model with 12.53B active parameters per token. It is built from four 12B experts for coding, agentic work, systems, and science. The model is released under Apache‑2.0 and supports only text generation.

Good at

  • Coding and tool use
  • Structured agent planning
  • Scientific reasoning

Limits

  • No vision support
  • Requires large VRAM (≈27.6 GB at 8 k context)
  • Sparse activation does not reduce download size

Best for

  • Local agentic workflows
  • Complex code generation
  • Scientific analysis

Who it’s for. Developers and researchers who need a local, text‑only model with expert routing for coding, planning, and scientific tasks. The model is experimental and not an official Google release.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—0.0 GB 13.4 GB
Too big Get file →
IQ1_S8.2 GB 21.6 GB
Too big Get file →
IQ1_M9.0 GB 22.4 GB
Too big Get file →
IQ2_XXS10.4 GB 23.8 GB
Too big Get file →
IQ2_XS11.5 GB 24.9 GB
Too big Get file →
IQ2_S11.7 GB 25.1 GB
Too big Get file →
IQ2_M12.8 GB 26.2 GB
Too big Get file →
Q2_K_S13.3 GB 26.7 GB
Too big Get file →
Q2_K14.3 GB 27.6 GB
Too big Get file →
IQ3_XXS14.8 GB 28.2 GB
Too big Get file →
IQ3_XS15.9 GB 29.2 GB
Too big Get file →
IQ3_S16.7 GB 30.1 GB
Too big Get file →
Q3_K_S16.7 GB 30.1 GB
Too big Get file →
IQ3_M17.1 GB 30.5 GB
Too big Get file →
Q3_K_M18.5 GB 31.9 GB
Too big Get file →
Q3_K_L20.0 GB 33.4 GB
Too big Get file →
IQ4_XS20.5 GB 33.9 GB
Too big Get file →
Q4_021.7 GB 35.1 GB
Too big Get file →
Q4_K_S21.8 GB 35.2 GB
Too big Get file →
Q4_K_M23.2 GB 36.5 GB
Too big Get file →
Q4_123.9 GB 37.3 GB
Too big Get file →
Q5_K_S26.3 GB 39.6 GB
Too big Get file →
Q5_K_M27.0 GB 40.4 GB
Too big Get file →
Q6_K31.2 GB 44.5 GB
Too big Get file →

For 8 k context, Q4_K_M (file size 14.26 GB, memory 27.6 GB) is a good balance. If VRAM is limited to about 22 GB, IQ1_S (8.2 GB, 21.6 GB) is the smallest viable option. Larger quants such as Q5_K_M provide higher fidelity at the cost of more memory.

What GPU do you need for Caelum G4 38B A12.5B?

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: the best quant that fits is the main file (0.0 GB file, about 13.4 GB in use).
  • 24 GB VRAM: the best quant that fits is IQ2_XXS (10.4 GB file, about 23.8 GB in use).
  • 32 GB VRAM: the best quant that fits is Q3_K_M (18.5 GB file, about 31.9 GB in use).
  • 48 GB VRAM: the best quant that fits is Q6_K (31.2 GB file, about 44.5 GB in use).

Which Caelum G4 38B A12.5B quant should you choose?

Q4_K_M is the usual starting point for Caelum G4 38B A12.5B: it balances file size and quality well. Here it is a 23.2 GB file.

If memory is tight, the smallest file is the first one in the table at 0.0 GB, with the quality trade-offs that come with fewer bits.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturegemma4
Parameters38B (approximate, mixture of experts)
Active parameters12.5B
Layers48
Trained context262,144 tokens
KV cache (fp16)1536 KB per token

Source

Quantized bymradermacher
FamilyCaelum
Licenseapache-2.0
Downloads (HF)873
Last updated2026-08-09

Frequently asked questions about Caelum G4 38B A12.5B

How much VRAM does Caelum G4 38B A12.5B need?

About 36.5 GB at Q4_K_M with an 8K context window. That is the 23.2 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 Caelum G4 38B A12.5B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 13.4 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Caelum G4 38B A12.5B GGUF?

It depends on the quant: file is 0.0 GB, IQ1_S is 8.2 GB, IQ1_M is 9.0 GB, IQ2_XXS is 10.4 GB, IQ2_XS is 11.5 GB, IQ2_S is 11.7 GB and 18 more.

What context length does Caelum G4 38B A12.5B support?

The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 12.9 GB at 8K and 51.5 GB at 32K.

Where can I download Caelum G4 38B A12.5B GGUF?

The files are on Hugging Face in the mradermacher/Caelum-G4-38B-A12.5B-i1-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 Caelum G4 38B A12.5B 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 this model on 8 GB VRAM?

No, the smallest quant requires at least 21.6 GB of VRAM for 8 k context.

Does it support vision inputs?

No, the model is text‑only; vision support is not released.

Can I use it with llama.cpp?

Yes, all GGUF files are llama.cpp‑compatible.

Other GGUFs of Caelum-G4-38B-A12.5B

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Caelum models

All Caelum →

Other large models

All large models →