Nex-N2.5 Guide: VRAM Requirements, GGUF Builds
About This Model
Nex-N2.5 is a family of models built specifically for agentic work, released by Nex-AGI. Its target is “long-horizon tasks in real-world environments": operating a computer or browser, writing, running and testing programs, and correcting itself along the way by looking at the screen to check results.
It comes in three sizes:
| Model | Size | Input | Role |
|---|---|---|---|
| Nex-N2.5-mini | ~35B MoE | Text and images | Small enough to run locally |
| Nex-N2.5-Pro | ~397B MoE | Text and images | Performance-focused |
| Nex-N2.5-Max | ~1.6T MoE | Text only | Flagship (not yet covered on this site) |
mini and Pro build on the multimodal (image-reading) foundation of the previous Nex-N2 generation, with focused improvements in computer use, browser use and visually grounded agent behavior. In the published config files, the architecture is Qwen3.5’s MoE (qwen3_5_moe).
What Makes It Stand Out
In the publisher’s comparison, Pro comes within a few points of closed frontier models on computer use, terminal use and web research, while a large gap remains in software engineering:
→ Scroll horizontally to see all columns
| Benchmark (what it measures) | mini | Pro | Max | Claude Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|---|
| OSWorld-Verified (operating a PC via the screen) | 71.2 | 82.2 | — | 83.4 | 83.2 |
| Terminal-Bench 2.1 (terminal use) | 73.4 | 82.7 | 86.1 | 89.1 | 88.8 |
| SWE-Bench Pro (real software engineering) | 43.8 | 61.2 | 65.7 | 79.2 | 64.6 |
| BrowseComp (web research) | 83.4 | 89.7 | 92.6 | 90.8 | 90.4 |
- On computer use (OSWorld-Verified), Pro is nearly level with frontier models. (The highest score in the same table is Qwen3.8-Max at 86.1.)
- mini scores in the 70s at a size you can run locally. For a ~35B MoE, 71.2 on OSWorld-Verified and 73.4 on Terminal-Bench 2.1 make it a strong candidate as the brain of an agent on your own machine.
- The gap in software engineering is large: on SWE-Bench Pro, Pro scores 61.2 versus 79.2 for Claude Opus 5; on DeepSWE v1.1 in the same card, 55.8 versus 73.7; and on the Job Bench work-task benchmark, 41.4 versus 65.7.
One caveat: the coding and computer-use results were measured with the publisher’s own harnesses (NexAU and NexCUA). Agent scores vary a lot with the harness, so these are not strictly like-for-like with other developers’ published numbers.
Running It Locally
- mini is the realistic choice. With the GGUF and other converted builds this site tracks (see “Quantized and Converted Variants" below), you can try it on common GPUs depending on the quantization level. Pro needs about 98GB of memory even at the smallest 1-bit-class quantization (IQ1_S), by this site’s estimate.
- The publisher recommends serving with its patched SGLang Docker image (
nexagi/sglang:v0.5.18-nex-patch). - If you want to try it without downloading the weights, hosted versions of mini and Pro are available through OpenRouter.
- The license is Apache-2.0, which allows commercial use.
Sources: model cards and config files for nex-agi/Nex-N2.5-mini and nex-agi/Nex-N2.5-Pro, as of 2026-09-25. Benchmark figures are as published by Nex-AGI.
Our Coverage and Data
Everything Local Model Watch has published about the Nex-N2.5 family: 2 article(s) covering the base model and its fine-tunes, plus converted builds we tracked after publication. Memory requirements below are computed by this site from file sizes, not quoted from model cards. Part of our model family index.
At a Glance
| Item | Value |
|---|---|
| Base model(s) | nex-agi/Nex-N2.5-Pro, nex-agi/Nex-N2.5-mini |
| Publisher | Nex-AGI |
| License (model card) | apache-2.0 |
| Articles | 2 |
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 |
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.
Figures are for nex-agi/Nex-N2.5-Pro. Other sizes in this family are covered in the articles below.
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.
This assessment is for nex-agi/Nex-N2.5-Pro.
Quantized and Converted Variants
→ Scroll horizontally to see all columns
| Added | Publisher | Format | Repository | Smallest VRAM tier (build, est. memory) |
|---|---|---|---|---|
| 2026-09-18 | bartowski | GGUF (imatrix) | bartowski/Nex-N2.5-Pro-GGUF | IQ1_S 98.2GB (does not fit a single consumer GPU) |
| 2026-09-18 | DevQuasar | GGUF | DevQuasar/nex-agi.Nex-N2.5-Pro-GGUF | Q2_K 161.8GB (does not fit a single consumer GPU) |
| 2026-09-18 | bartowski | GGUF (imatrix) | bartowski/nex-agi_Nex-N2.5-mini-GGUF | Q2_K 15.5GB (fits in 16GB VRAM) |
| 2026-09-18 | mradermacher | GGUF | mradermacher/Nex-N2.5-mini-GGUF | Q2_K 14.5GB (fits in 16GB VRAM) |
| 2026-09-18 | mlx-community | MLX | mlx-community/Nex-N2.5-mini-OptiQ-4bit | MLX 4bit 24.8GB (fits in 32GB VRAM) |
| 2026-09-18 | mlx-community | MLX | mlx-community/Nex-N2.5-mini-oQ4 | MLX 4bit 22.8GB (fits in 24GB VRAM) |
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
- Available builds in bartowski/nex-agi_Nex-N2.5-mini-GGUF: IQ2_XXS 10.1GB / IQ2_XS 10.5GB / IQ2_S 11.2GB / IQ2_M 12.3GB / Q2_K 12.9GB / IQ3_XXS 14.5GB / IQ3_XS 15.2GB / Q3_K_S 15.2GB / Q3_K_M 16.2GB / Q3_K_L 17.0GB / IQ3_M 17.7GB / IQ4_XS 18.5GB / Q4_0 18.6GB / Q4_K_S 19.5GB / Q4_1 20.5GB / IQ4_NL 20.8GB / Q4_K_M 20.8GB / Q4_K_L 22.5GB / Q5_K_S 23.4GB / Q5_K_M 25.1GB / Q6_K 27.4GB / Q6_K_L 30.1GB / Q8_0 34.4GB / BF16 64.6GB
- Available builds in mradermacher/Nex-N2.5-mini-GGUF: Q2_K 12.1GB / Q3_K_S 14.1GB / Q3_K_M 15.6GB / Q3_K_L 16.9GB / IQ4_XS 17.6GB / Q4_K_S 18.5GB / Q4_K_M 19.7GB / Q5_K_S 22.3GB / Q5_K_M 23.0GB / Q6_K 26.6GB / Q8_0 34.4GB
- Available builds in mlx-community/Nex-N2.5-mini-OptiQ-4bit: MLX 4bit 20.6GB
- Available builds in mlx-community/Nex-N2.5-mini-oQ4: MLX 4bit 19.0GB
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.
Articles (the family’s own models first, then newest)
| Published | Model | Type | Article |
|---|---|---|---|
| 2026-09-09 | nex-agi/Nex-N2.5-Pro | New Models | Nex-N2.5-Pro Long-Horizon Agent Model: ~98GB Memory, GGUF Builds |
| 2026-09-09 | nex-agi/Nex-N2.5-mini | New Models | Nex-N2.5-mini Agent Model for Long-Horizon Tasks: 16GB+ VRAM |
Repositories
Last updated 2026-09-28 (JST). The explanation at the top of this page was written with the help of AI from the primary sources it cites. The tables and lists under “Our Coverage and Data" are assembled by code from our article log.