Breaking

Nex-N2-Pro Lands 12th in Design Arena's Open-Weight Rankings at Elo 1275

June 24, 2026 at 15:04 EDT

Foundation Models
Open Source
AI Agents

Nex AGI's open-weight model Nex-N2-Pro has placed 12th (Elo 1275) in the open-weight category of Design Arena, a benchmark that measures design and app-generation ability. In the 3D Design Arena category it ranked higher, at 16th (Elo 1314).

June 2026 · Nex AGI

Open-Weight "Nex-N2-Pro" Breaks Into Design Arena's Upper Ranks

The Apache-2.0 MoE model lands 12th in the open-weights category and 16th in 3D — sitting in the same performance band as MiniMax M2.7 and GLM 4.7, with frontier-level claims in coding and agentic workflows.

#12

Open-weights category
Elo 1275

#16

3D Design Arena
Elo 1314

397B

Total params
17B active · MoE

262K

Context window
text + image input

Published Benchmark Scores

Column height drawn proportional to each reported score (0–100 scale).

90.7

83.7

80.8

75.3

58.8

33.6

GPQA
Diamond

Browse
Comp

SWE-Bench
Verified

Terminal-
Bench 2.1

SWE-Bench
Pro

DeepSWE

"Agentic Thinking" framework

Turning thinking into action across long-horizon agent tasks — built on three integrated steps.

Reasoning

→

Tool Use

→

Environment Execution

Where it shines

Strong on coding & agent-style workflows
Consistent in local execution
Runs stably even on a 128GB Mac
Frontier-level claims vs GPT-5.5 / Claude Opus 4.7

Where it lags

Weak as a pure chatbot — over-replies to a simple "hi"
Early tool-calling issues reported
Below frontier on some benchmarks (e.g. DeepSWE 33.6)
Full-precision needs hundreds of GB VRAM — quantize

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.