Nex-N2.5-Pro Long-Horizon Agent Model: ~98GB Memory, GGUF Builds

November 3, 2026

Nex-N2.5-Pro Long-Horizon Agent Model: ~98GB Memory, GGUF Builds

At a Glance

Item Value
Repository nex-agi/Nex-N2.5-Pro
Family guide Nex-N2.5 guide (2 articles)
Publisher guide Nex-AGI: models and licenses
Published 2026-09-08
License apache-2.0
Formats safetensors
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

What we checked ourselves

  • It needs about 21% fewer tokens than the Qwen3 tokenizer for the same Japanese text, so about 1.26× as much Japanese fits in the same context length, and generation is faster per character.

Details and conditions are in “Our Own Measurements” below.

Overview

Nex-AGI has announced “Nex-N2.5", a family of agent models designed for real-world long-horizon tasks. Available in three sizes—mini, Pro, and Max—the model weights are open-sourced on platforms like Hugging Face. A key feature is the enhanced continuous action and self-correction capabilities driven by visual feedback.

Specifications

  • License: apache-2.0
  • Library: transformers
  • Pipeline task: text-generation

Performance

Evaluation results, based on model card descriptions and publisher measurements, include both text and multimodal benchmarks. Scores for Nex-N2.5-mini, Nex-N2.5-Pro, and Nex-N2.5-Max are listed alongside major competitor models for comparison.

Text Benchmarks

→ Scroll horizontally to see all columns

Benchmark / CODING 3 Nex-N2.5-mini / CODING 3 Nex-N2.5-Pro / CODING 3 Nex-N2.5-Max / CODING 3 Claude Opus 5 / CODING 3 GPT-5.6 Sol / CODING 3
Terminal-Bench 2.1 73.4 82.7 86.1 89.1 88.8
SWE-Bench Pro 43.8 61.2 65.7 79.2 64.6
DeepSWE v1.1 36.1 55.8 65.6 73.7 72.7
AGENTIC AGENTIC AGENTIC AGENTIC AGENTIC AGENTIC
AutomationBench v1.0.6 5 32.3 44.2 50.2 50.3 45.8
Toolathlon Verified 54.6 68.5 74.7 76.5 74.9
GDPval-AA v2 1446 1628 1713 1831 1711
Job Bench 28.5 41.4 53.6 65.7 45.4
BrowseComp 6 83.4 89.7 92.6 90.8 90.4

Multimodal Benchmarks

→ Scroll horizontally to see all columns

Benchmark Nex-N2.5-mini Nex-N2.5-Pro MiniMax-M3 Claude Opus 5 GPT-5.6 Sol
OSWorld-Verified 8 71.2 82.2 75.2 83.4 83.2
OSWorld-2 30.5 56.4 22.3 68.3 62.7
WebTest 8, 9 48.6 52.8 — — 54.0
WebArena-Verified 8 63.4 67.6 — — 69.7
OSWorld-G 82.9 87.4 — 76.8 77.7
Vision2Web 7 52.9 68.2 59.0 — 79.8
SWE-MM 25.5 38.2 — 59.4 40.2
OmniDoc 89.7 92.2 91.6 — 92.9

According to evaluations by the publishers, Nex-N2.5-Pro achieves 82.7 on Terminal-Bench 2.1 and 61.2 on SWE-Bench Pro. It also records a score of 82.2 on OSWorld-Verified. While it demonstrates solid performance in terminal operation and agent-related tasks, it trails behind the top-tier Max model and certain comparison models in some metrics. Note that the reasoning_effort parameter can be used to control the thinking mode.

Strengths and Use Cases

  • Computer operation, web browsing, and visually grounded agent capabilities (computer use)
  • Long-horizon tasks and autonomous program execution/testing
  • Scientific research, knowledge work, and complex productivity tasks
  • Function calling and utilization of reasoning parsers

Hardware Requirements

Estimated requirements (calculated by Local Model Watch) — 396.8B parameters

Your VRAM Quantization File size Est. memory needed
More than 98GB of VRAM (multi-GPU or CPU offload required) IQ1_S 81.8GB 98.2GB

Inference engine support (architecture name matched against each project’s own model registry in its source code, checked 2026-11-03): llama.cpp: registered, vLLM: registered, MLX (mlx-lm): registered.

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. File sizes are measured from the converted build bartowski/Nex-N2.5-Pro-GGUF. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

Runs in Ollama, LM Studio and llama.cpp via a converted build.

The publisher ships safetensors, but bartowski/Nex-N2.5-Pro-GGUF provides a GGUF build you can use.

License — apache-2.0 (Commercial use allowed): Permits commercial use, modification and redistribution. Redistribution requires including the license and stating changes; includes a patent grant.

Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists. License summaries are not legal advice — check the publisher’s original terms before relying on them.

Our Own Measurements

Values we measured ourselves on our server (no GPU) by actually reading and running this model’s files — not figures copied from the model card. How we measure, and the results for every model: Our Measurements.

Japanese Token Efficiency

Tokenizer Tokens per 1,000 Japanese characters Ratio to the same text in English
This model 546 0.99×
Qwen3 688 1.26×
Llama 3.2 744 1.36×
Gemma 3 564 1.03×
gpt-oss 795 1.45×
LLM-jp-3 497 0.85×

It needs about 21% fewer tokens than the Qwen3 tokenizer for the same Japanese text, so about 1.26× as much Japanese fits in the same context length, and generation is faster per character.

Counted with the tokenizer.json of nex-agi/Nex-N2.5-Pro on a fixed text we wrote ourselves (876 Japanese characters across news, conversation, technical docs, a formal email, travel writing and a recipe) and its English translation. Fewer tokens mean more Japanese fits in the context window.

Inside the GGUF File

File: Nex-N2.5-Pro-Q4_K_M/Nex-N2.5-Pro-Q4_K_M-00001-of-00007.gguf (233.61GB, Q4_K_M). We read only the header (metadata) of the file, not the weights.

Item Value
Architecture (as named in the GGUF) qwen35moe
Maximum trained context length 262,144 tokens
Layers 60
Experts 10 active out of 512
Vocabulary size 248,320
Chat template Included (mentions tool calls, has a thinking switch)
imatrix Used (Nex-N2.5-Pro-calibration-v6.txt, 573 chunks)
Weight types (share of parameters) Q4_K 78.0% / Q6_K 16.8% / Q8_0 5.1% / F32 0.0%
Average bits per weight 5.06 bits
Embedding / output layer type Q8_0 / Q8_0

The quant name in the file name describes the file as a whole; in practice layers mix several types. The average bits per weight is the measured file data divided by the number of weights.

Recent Models in the Same Size Class

Models with over 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
deepseek-ai/DeepSeek-V4-Pro-0813 1650.5B — mit DeepSeek-V4-Pro-0813 Text Generation Model: ~998GB Memory, GGUF Builds (2026-09-26)
nvidia/DeepSeek-V4-Pro-0813-nvfp4-DSpark 1650.5B — mit DeepSeek-V4-Pro-0813-nvfp4-DSpark: ~1005GB Memory (2026-09-10)

How to Get It

  • Distribution: Model weights are available on platforms such as Hugging Face and ModelScope.
  • Supported Engines & Deployment: Deployment via SGLang (using a custom nexagi/sglang:v0.5.18-nex-patch Docker image) is provided, and can be launched from the command line.

Quantized and Converted Variants

→ Scroll horizontally to see all columns

Added Publisher Format Repository Smallest VRAM tier (build, est. memory)
2026-09-18 (9 days after publication) bartowski GGUF (imatrix) bartowski/Nex-N2.5-Pro-GGUF IQ1_S 98.2GB (does not fit a single consumer GPU)
2026-09-18 (9 days after publication) DevQuasar GGUF DevQuasar/nex-agi.Nex-N2.5-Pro-GGUF Q2_K 161.8GB (does not fit a single consumer GPU)

File sizes of each build:

  • Available builds in bartowski/Nex-N2.5-Pro-GGUF: IQ1_S 81.8GB / IQ1_M 91.3GB / IQ2_XXS 107.2GB / IQ2_XS 113.1GB / IQ2_S 119.0GB / IQ2_M 132.5GB / Q2_K 140.4GB / IQ3_XXS 157.9GB / IQ3_XS 165.9GB / Q3_K_S 167.7GB / Q3_K_M 176.9GB / Q3_K_L 186.5GB / IQ3_M 195.1GB / IQ4_XS 204.5GB / Q4_0 210.0GB / Q4_K_S 217.5GB / Q4_1 232.0GB / IQ4_NL 232.4GB / Q4_K_M 233.6GB / Q5_K_S 263.4GB / Q5_K_M 284.6GB / Q6_K_S 310.3GB / Q6_K 325.8GB / Q8_0 392.6GB
  • Available builds in DevQuasar/nex-agi.Nex-N2.5-Pro-GGUF: Q2_K 134.8GB / Q3_K_M 176.5GB / Q4_K_M 224.1GB / Q5_K_M 262.4GB

In addition, 2 converted build(s) from other uploaders exist on Hugging Face; this site lists only builds from the model’s publisher or established quantization maintainers.

This section is appended automatically by Local Model Watch when a converted build of this model appears after publication. Memory figures are estimated from the size of the distributed files. What the quantization names mean: glossary.

Related Articles

What to Read Next

Sources

Update History

  • 2026-09-18: Added converted builds to “Quantized and Converted Variants”: bartowski/Nex-N2.5-Pro-GGUF, DevQuasar/nex-agi.Nex-N2.5-Pro-GGUF
  • 2026-09-18: Updated the hardware requirements table with the actual file sizes of bartowski/Nex-N2.5-Pro-GGUF.
  • 2026-09-24: Rebuilt the article: Added a “can you run it locally?" verdict from the distribution formats, converted builds, engine registries, license terms and measured compression
  • 2026-09-25: Changed the title to show what the article covers (VRAM requirements, file list, etc.).
  • 2026-09-28: Added our own measurements: Japanese token efficiency, what is inside the GGUF file.

New ModelsAgent,Hugging Face,Nex-AGI,Nex-N2.5-Pro,SGLang,確認済み