whatmodel.sh / Nex-N2.6
Nex-N2.6
↗ share
+ contribute
Nex-N2.6 is an open-weights language model from Nex-AGI, released November 2026. Its published size is 321B parameters. Licence: apache-2.0.
Catalogue identity notes
Nex-AGI fine-tune of GLM-5.3-Flash (MoE), successor to Nex-N2.5-Pro, focused on computer use, coding and long-horizon tool use; image input. Repo (created 2026-11-10) holds FP8 weights (328.3 GB); card results are 'model-level results, not a separate evaluation of this FP8 export'. The card's self-hosting route is the GLM-5.3-Flash SGLang guide with the model path replaced (no copy-pasteable command), so local routes use bartowski's GGUF quants. reasoning_effort low/high/max, default max. No official Ollama entry (404). Card states no context length. Licence: No LICENSE file in the repo at 9874722; licence from the card metadata 'license: apache-2.0'. Base model GLM-5.3-Flash is MIT per the DB. API price: No first-party price; OpenRouter prices are not an allowed price source.
Model information sources
Release date ↗ · Weights licence ↗ · Parameter count ↗
Card: 'approximately 321B parameters' (FP8 export). The card does not state active parameters; it says the model is 'built on GLM-5.3-Flash', whose pinned card states '320B total parameters and just 18B active parameters' (https://huggingface.co/zai-org/GLM-5.3-Flash/blob/eb9eb208eb0d988989d07a6a12d0fdeb5f52574a/README.md); active parameters are not recorded because the Nex card itself does not state them. HF safetensors metadata: 321,323,031,390 (F8_E4M3 314,396,639,232; BF16 6,922,754,304; F32 3,637,854).
$ how it compares
Scored LLMs released closest to Nex-N2.6, out of 1206 with a score.
| # | model | score | where |
|---|---|---|---|
| 1 | 89±1 | api $16.00/M | |
| 14 | 79±2 | api $3.00/M | |
| 22 | 78±1 | api price unknown | |
| 32 | 76±1 | api $0.20/M | |
| 37 | 75±1 | api price unknown |
$ what it's good at
agent tasksnot ranked on this task yet · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Terminal-Bench 2.1 lab | 84.1% ✎ | ▁·▁·▁▁▄▃▅█▆█ beats 53/68 | — |
$ where to run it
Through an API
api · OpenRouter (the API link on the Nex-AGI card) · price unknown · openrouter.ai ↗
Memory is what a version takes in GPU memory (or a Mac's unified memory) with an 8k-token conversation. With less, it runs partly from system RAM, more slowly. Enter your machine to see what runs at full speed.
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run Nex-N2.6 · needs 214 GB memory, 202 GB download · from huggingface.co/bartowskichat in the terminal
llama cli -hf bartowski/Nex-N2.6-GGUF:Q4_K_M - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf bartowski/Nex-N2.6-GGUF:Q4_K_M
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run Nex-N2.6 · needs 214 GB memory, 202 GB download · from huggingface.co/bartowski
- Or download it in the terminal
lms get https://huggingface.co/bartowski/Nex-N2.6-GGUF@Q4_K_M -y - Then run lms ls to see its name, and lms chat followed by that name
Model repository and documentation: huggingface.co/nex-agi ↗
$ published results · 1 result, not enough overlap for a score
| test | result | source and conditions |
|---|---|---|
| Terminal-Bench 2.1 | 84.1% | nex-agi/Nex-N2.6 model card · maker-reported Nex-AGI own result in the pinned nex-agi/Nex-N2.6 card (sha 9874722), Performance table, row 'Terminal-Bench 2.1', last column 'Nex-N2.6' = 84.1. Card: 'The following results are reported by the Nex-AGI team for this release. Scores depend on the evaluation harness and settings. These are model-level results, not a separate evaluation of this FP8 export.' Harness and reasoning effort not stated (template default effort is max). Other columns not used. |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json