Pantheon-Reasoning-26B-A4B-1.1-V2 GGUF Quantizations
At a Glance
| Item | Value |
|---|---|
| Repository | bartowski/Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-GGUF |
| Published | 2026-09-10 |
| License | apache-2.0 |
| Formats | GGUF |
| Source type | Primary source (the publisher itself) |
Values determined by this site’s code at collection time. Dates are JST.
Overview
[Important] The information introduced in this article has not yet been officially confirmed and is based on breaking data at the time of writing. Please note that details may change due to future verification or official announcements.
It is reported that the prominent uploader bartowski has released a GGUF quantized version of Gryphe’s model, bartowski/Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-GGUF. This is a 26B model specialized for roleplay and reasoning based on the Gemma 4 architecture, and various quantized files are provided for local execution.
Specifications
- License: apache-2.0
- Base Model: Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2
- Architecture: Gemma4ForConditionalGeneration
Performance
According to the base model’s model card and published information, comparison data regarding text quality and reasoning trace behavior during roleplay has been shown. As for writing quality trends across different comparison targets, for example, the Type-token ratio is 0.770 for Pantheon Reasoning and 0.748 for Pantheon Reasoning (no-think), shifting from 0.686 for the base model.
Additionally, regarding the frequency of cliché occurrences, it is reported that while Base instruct has 1.141 occurrences per 100 words, Pantheon Reasoning sees a significant reduction to 0.546 occurrences, and Pantheon Reasoning (no-think) to 0.541 occurrences. On the other hand, since enabling the thinking process outputs reasoning traces, users should note that there is a trade-off regarding response length and latency depending on the task.
Strengths and Use Cases
The base model is reported to be a model specialized for roleplay and reasoning that adopts the Gemma 4 MoE architecture. It has been fine-tuned with advanced creative writing and roleplay use cases in mind, such as character psychological description, tone adjustment, and narrative development planning, and is trained so that the thinking process is enabled across all reasoning turns.
Hardware Requirements
Estimated requirements (calculated by Local Model Watch) — 26.5B parameters (taken from the base model Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2)
| Your VRAM | Quantization | File size | Est. memory needed |
|---|---|---|---|
| 12GB (RTX 4070 / 3060 12GB, etc.) | IQ2_S | 10.0GB | 12.0GB |
| 16GB (RTX 5060 Ti 16GB / 4060 Ti 16GB, etc.) | IQ3_XS | 12.7GB | 15.2GB |
| 24GB (RTX 4090 / 3090, etc.) | Q5_K_M | 19.4GB | 23.2GB |
| 32GB (RTX 5090, etc.) | Q8_0 | 25.7GB | 30.9GB |
| 80GB class (A100 / H100) | BF16 | 48.4GB | 58.1GB |
Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.
Recent Models in the Same Size Class
Models with 15–40B parameters that Local Model Watch covered recently, listed by code from our article log for comparison. VRAM tiers are this site’s estimates; licenses are as stated on the model cards.
→ Scroll horizontally to see all columns
| Model | Parameters | Smallest VRAM tier | License | Our article |
|---|---|---|---|---|
| Edge0/Edge0-35B-A3B-preview | 34.7B | 80GB | apache-2.0 | Edge0-35B-A3B-Preview: Sparse MoE for Phone-Class Memory (2026-09-11) |
| nex-agi/Nex-N2.5-mini | 35.1B | 16GB | apache-2.0 | Nex-AGI Releases Open-Weight Long-Task Model Nex-N2.5-mini (2026-09-09) |
How to Get It
It is distributed in GGUF format and can be downloaded using the Hugging Face CLI or similar tools. It is reported to work with engines such as llama.cpp (release b10842 or later recommended), LM Studio, koboldcpp, ramalama, Jan AI, Text Generation Web UI, LoLLMs, and Atomic Chat.
Example download command:
hf download bartowski/Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-GGUF --include "Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-Q4_K_M.gguf" --local-dir./
Related Articles
- Salience-27B-R6 GGUF Released by bartowski
- Intern-S2-397B GGUF Quantized Models Released by bartowski
- Orion-26B-A4B-v1.1 GGUF Quants Released by bartowski
- bartowski/nex-agi_Nex-N2.5-mini-GGUF: Specs and Hardware