Skip to content
N

Nex-N2-Pro

Open Source

Nex AGIReleased on 2026-06-02

Nex-N2-Pro is an open-weights agentic mixture-of-experts model from Nex AGI, with 17B active parameters out of 397B total, built on the Qwen3.5 architecture. It accepts text and image input and is tuned for long-horizon agentic work, frontier coding, and tool use, with a 262K-token context window. Released and open-sourced under Apache 2.0 on 2026-06-02. Reported benchmarks include SWE-Bench Verified 80.8, Terminal-Bench 2.1 75.3, GPQA Diamond 90.7, and BrowseComp 83.7 — strong among open-weights models, though it trails closed frontier models (GPT-5.5, Claude Opus 4.7) on most coding suites.

82
Overall Score

Voice of the community

New release: Nex-N2-Pro, a 397B parameter MoE model based on Qwen — open-sourced under Apache 2.0 with first-tier agentic and coding capabilities.

NVIDIA Developer Forums2026-06-05

Nex-N2-Pro brings agentic thinking, frontier coding, and long-horizon execution to open weights.

SiliconFlow2026-06-03

Core Specs

262K
Context Window
262K
Max Output
ReasoningOpen Sourcetextimage

Pros & Cons

Sentiment100% +0% ·0% −

Pros

  • +Open weights under permissive Apache 2.0 license — self-hostable and commercially usable
  • +Strong agentic & coding benchmarks for an open model (SWE-Bench Verified 80.8, Terminal-Bench 2.1 75.3)
  • +262K-token context with native text + image input
  • +Efficient MoE: only 17B of 397B parameters active per token; free via OpenRouter and multiple hosted providers

Cons

  • Trails closed frontier models (GPT-5.5, Opus 4.7) on most coding suites (SWE-Bench Pro 58.8 vs Opus 4.7 64.3)
  • 397B total parameters make self-hosting hardware-intensive
  • New provider with limited Western ecosystem, tooling, and independent third-party benchmarks
  • No first-party hosted API with SLA — relies on third-party inference providers

Pricing

Input (per 1M tokens)$0.00
Output (per 1M tokens)$0.00
Free trial available
Updated on 2026-06-09

Get Started

1Visit the provider's website
2Create an account
3Start using the model

Benchmarks

userRating%
aiIndex%
sweBenchVerified80.8%
terminalBench2175.3%
sweBenchPro58.8%
gpqaDiamond90.7%
browseComp83.7%
ifeval94%

Reliability

Incidents (30d)0