Model library / Gryphe
Pantheon-Reasoning-26B-A4B-1.1-V2 is a 26.5 billion-parameter Gemma-4 mixture-of-experts finetune aimed at role-play scenarios, incorporating full reasoning traces to improve character planning and narrative coherence.
Mixture of experts, count not published
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
Pantheon-Reasoning-26B-A4B-1.1-V2 is a 26.5 billion-parameter Gemma-4 mixture-of-experts model released by Gryphe. It is fine-tuned for role-play applications, using full reasoning traces to help the model plan character responses and narrative direction.
Deployers receive a 26.5 billion-parameter model stored in bfloat16, licensed under Apache-2.0, with a 262 k token context window. It uses a Gemma-4 MoE architecture, though the exact expert count is not disclosed. Training mixes roughly 26 % Pantheon role-play, 36 % general role-play and 38 % text-adventure data, all augmented with back-generated reasoning traces.
Specification
| Repository | Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 |
| Publisher | Gryphe |
| Published | 8 September 2026 |
| Architecture | Gemma4ForConditionalGeneration |
| Model type | gemma4 |
| Parameters | 26,544,131,376 |
| Layers | 30 |
| Hidden size | 2,816 |
| Attention heads | 16, 8 key/value heads |
| Context length | 262,144 tokens |
| Vocabulary | 262,144 tokens |
| Weight format | bfloat16 |
| Licence | apache-2.0 |
| Base model | Gryphe/Gemma-4-26B-A4B-StyleTune-V2 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 49.4 GiB | Full precision as released |
| fp8 | 24.7 GiB | 8-bit, near-lossless on most models |
| int4 | 12.4 GiB | 4-bit, smallest footprint |
The cache costs 240 KiB per token at 16-bit, so the full 262,144-token context of one request needs about 60.0 GiB. Concurrency multiplies that number, not the weights.
That figure is an upper bound: this model uses sliding-window attention with a 1,024-token window on some layers, so long requests cache less than this.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 4 | 27.0 GiB | about 117,779 tokens |
| RTX 5090 32GB | 2 | 3.2 GiB | about 13,796 tokens |
| L40S 48GB | 2 | 32.0 GiB | about 139,625 tokens |
| A100 80GB | 1 | 20.1 GiB | about 87,633 tokens |
| H100 80GB | 1 | 20.1 GiB | about 87,633 tokens |
| RTX PRO 6000 96GB | 1 | 34.5 GiB | about 150,547 tokens |
| H200 141GB | 1 | 75.0 GiB | the full 262,144 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve Gryphe/Pantheon-Reasoning-26B-A4B-1.1-V2 \ --max-model-len 114688 \ --gpu-memory-utilization 0.90
Questions
The weights take about 49 GiB at the released precision. That fits on one A100 80GB, which leaves roughly 20 GiB for the key-value cache.
The repository declares apache-2.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 262,144 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
Nex-N2.5-mini is a 35.1 billion-parameter mixture-of-experts language model released under Apache-2.0.
LLaDA2.2-mini is a 16.3 billion-parameter MoE diffusion language model with 128 k token context, supporting Levenshtein editing.
K2-Horizon-32B-Stage1 is a 32-billion-parameter dense decoder-only model released by IFM for benchmarking agentic, coding and rea…
K2-Horizon-MoVA-36B-A4B is a sparse Mixture-of-Experts language model with MoVA attention, released by IFM.
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.