Nex AGI releases Nex-N2-Mini: open-source agentic MoE model with 262K context window
Nex AGI has released Nex-N2-Mini, an open-source agentic mixture-of-experts model with a 262K-token context window. The model accepts text and image inputs and is priced at $0.025 per 1M input tokens and $0.10 per 1M output tokens.
Nex AGI releases Nex-N2-Mini: open-source agentic MoE model with 262K context window
Nex AGI has released Nex-N2-Mini, an open-source agentic mixture-of-experts model with a 262K-token context window, now available through OpenRouter.
Specifications
The model is priced at $0.025 per 1M input tokens and $0.10 per 1M output tokens. It accepts both text and image inputs and produces text output, making it a multimodal model.
Nex-N2-Mini is described as the "smaller sibling in the Nex-N2 series," indicating a larger model exists or is planned in the same family.
Capabilities
According to Nex AGI, the model is built for:
- Coding tasks
- Tool use
- Deep research
- Long-horizon agentic workflows
- Native reasoning support
The 262K-token context window places it among models with extended context capabilities, though below the largest available windows from providers like Anthropic (Claude with 200K standard) and Google (Gemini 1.5 Pro with 2M tokens).
Architecture
The model uses a mixture-of-experts (MoE) architecture, a design pattern that activates only subsets of parameters for each inference, potentially improving efficiency compared to dense models of similar capability.
Nex AGI has made the model weights available as open-source, allowing researchers and developers to download and deploy the model independently.
Availability
The model is currently hosted exclusively through OpenRouter, which forwards requests directly to the provider without routing decisions. OpenRouter reports that prompt caching can reduce effective costs by 60-80% below list prices for workloads with repeated context.
The official release date is listed as June 24, 2026, though this appears to be an error in the source data given the current date.
What this means
Nex-N2-Mini enters a competitive space for coding and agentic models, with its 262K context window and MoE architecture potentially offering cost-performance advantages for long-context tasks. The open-source release allows developers to self-host, though the lack of disclosed benchmark scores makes direct capability comparisons difficult. The pricing sits in the mid-range: cheaper than frontier models like GPT-4 but more expensive than some open models. Whether the model delivers on its claimed agentic capabilities will depend on independent testing and real-world usage data.
Related Articles
Thinking Machines Lab releases Inkling 2 Mini: 26B-parameter looped MoE with 3B active and 1M-token context
Thinking Machines Lab has released Inkling 2 Mini, a multimodal Mixture-of-Experts model with about 26B physical total parameters and 3B active per token. It repeats its middle layers three times, supports a 1M-token context, and ships with a DSpark drafter. Hugging Face reports it leads 32 of 48 benchmarks against four comparable models.
Tencent releases Hy4 flagship weights: 770B-parameter MoE, 1,048,576-token context, conditional license
Tencent released downloadable weights for its flagship Hy4 on November 20, with a System Card. The multimodal MoE has a 770-billion-parameter language backbone, 49 billion active parameters per text token and a 1,048,576-token context window, under a license with commercial conditions for model-as-a-service operators.
Sarvam AI releases open-weight Sarvam-105B-1711, a 105B MoE with 10.3B active parameters
Sarvam AI released Sarvam-105B-1711 on November 22, 2026, an open-weight checkpoint aimed at customer support, voice agents and other high-volume enterprise work. It ships as BF16 weights for further training and a 4-bit NVFP4 version designed to run on one workstation GPU.
NVIDIA Releases Alpamayo 2 Edge, a 2B-Class Reasoning VLA Running on DRIVE AGX Thor at 160 to 190 ms
NVIDIA released Alpamayo 2 Edge, a 2B-class reasoning vision-language-action (VLA) model that plans driving trajectories and explains them. It runs on DRIVE AGX Thor at 160 to 190 ms pixel-to-trajectory latency, according to MarkTechPost.
Comments
Loading...