Nex-N2.5-Pro reads text and pictures and writes text back, and it is built for coding that looks at the result on screen: it explores a codebase, changes several files, runs commands, opens the app, uses a browser or a desktop and checks the outcome as a user would. It also suits testing screens, automating a computer and deep research.
Nex-N2-Mini
by Nex AGIReleased 24 June 2026
Nex-N2-Mini is an open-source agentic mixture-of-experts model from Nex AGI, the smaller sibling in the Nex-N2 series. It accepts text and image input and is built for coding, tool use,...
Reads about 260 pages · writes up to 230 pages
Parameters you can send
Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.
Shaping the reply
How long, how varied, where to stop.
max_tokenstemperaturetop_ptop_klogprobstop_logprobsReasoning
Let it think before answering — and see the thinking.
reasoninginclude_reasoningStructured output
Get JSON you can parse instead of prose.
response_formatstructured_outputsTools
Let the model call your functions.
toolstool_choiceQuestions about Nex-N2-Mini
Is Nex-N2-Mini still on sale?
No. Nex AGI has retired it. Other Nex AGI models are on sale.
What is Nex-N2-Mini best for?
The lab has not said.
Does Nex-N2-Mini work in Arabic, French or Darja?
Nex AGI names no languages for Nex-N2-Mini. Arabic, French and Darja: not stated.
Can it read PDFs, images or a whole book?
It reads text and images. Up to 262K tokens in one request, about 260 pages. It writes up to 177 000 words at once.
Is Nex-N2-Mini open source?
Yes, its weights are public. Open weights on Hugging Face
Who makes Nex-N2-Mini, and when was it released?
Nex AGI. Released 24 June 2026.
Other Nex AGI models, priced in dinars
See all Nex AGI modelsNex-N2.5-Mini reads text and pictures and writes text back, and it is built to finish a task and prove it worked. It explores a codebase, edits several files, runs commands, launches the app, drives a browser or a desktop and tests as a user would, then revises when the result is wrong. It keeps about 256 pages in mind.