DeepSeek V3.1 Nex N1
nex-agi/deepseek-v3.1-nex-n1DeepSeek V3.1 Nex N1 (nex-agi/deepseek-v3.1-nex-n1) is a deepseek_v3 671.1B-parameter model from Nex Agi with a 131,072-token context window and 163,840 max output tokens, priced at $0.14/1M input and $0.50/1M output tokens. Available via the haimaker.ai OpenAI-compatible API.
Overview
Deepseek V3.1 Nex N1 is a chat model by Nex Agi. It has 671.1B parameters. It supports a 131K token context window. Supports function calling.
Model Card
Nex-N1
Nex is a next-generation, full-stack agentic platform that brings foundation models, synthetic data pipelines, RL training, agent frameworks, and deployment tools together in one unified ecosystem.
DeepSeek-V3.1-Nex-N1 is the flagship release of the Nex-N1 series — a post-trained model designed to highlight agent autonomy, tool use, and real-world productivity.
We are committed to making it easier than ever to build and deploy AI agents by offering researchers and entrepreneurs a high-performance, reliable, and cost-effective "out-of-the-box" agent system.
Highlights
- Full spectrum model matrix: From 8B to 671B parameters, the Nex series covers everything from edge-friendly setups to frontier-scale deployments.
- Agent-focused performance: Demonstrates industry-leading results on programming, tool-use, web-search, and other multi-hop reasoning tasks.
- Production-ready utility: Excels at mini-app development, website authoring, slide creation, and immersive role-play—delivering immediate productivity
- End-to-end control: Developers can build the entire data-to-deployment loop on top of Nex, ensuring sovereignty while keeping costs predictable.
- Open ecosystem: Turnkey synthetic data pipelines, curated datasets, Nex-N1 checkpoints, the NexAU Agent framework, the EaaS MoE inference stack, and NexRL
Performance
Nex-N1 is evaluated on six representative agentic benchmarks (general + professional). The model consistently ranks at or near the top across tool-using, web-search, and coding-heavy evaluations, showing strong readiness for real-world agent workflows.
- Data points are sourced by default from the model’s official technical report or blog, as well as the benchmark’s official results. All other metrics were tested in strict compliance with the official standard evaluation framework.
- Results for Tau2-bench are derived via a weighted average.
- For SWE-verified-bench, test results are based on an internal scaffold built with OpenHands—using a 128k context length and 150 maximum steps—and represent the average of four runs.
- Terminal-Bench2 is evaluated using the official Terminus2 agent.
Nex-N1 provides various size models from 8B to 671B for different usage scenarios.
| Model | GAIA2 | τ2-Bench | SWE-bench Verified | Terminal-Bench2 | BaxBench | BFCL v4 |
| --- | --- | --- | --- | --- | --- | --- |
| DeepSeek-V3.1-Nex-N1 | 29.5 | 80.2 | 70.6 | 31.8 | 59.7 | 65.3 |
| Qwen3-32B-Nex-N1 | 16.7 | 72.1 | 50.5 | 16.7 | 34.8 | 60.5 |
| Qwen3-30B-A3B-Nex-N1 | 11.3 | 65.3 | 29.7 | 8.3 | 13.6 | 51.9 |
| internlm3-8B-Nex-N1 | 8.6 | 63.0 | 20.3 | - | - | 44.5 |
Nex-N1 demonstrates competitive performance across all evaluation scenarios, showing particularly strong results in practical coding and HTML generation tasks.
Refer to and for more details.
Usage
Local Deployment
We recommend sglang for serving Nex-series models locally:
python -m sglang.launch_server --model-path /path/to/your/modelFunction Calling
Nex-series models support robust function-calling capabilities. To maximize the function-calling capabilities of the Nex-series models, we modified the tool parser of qwen3_coder, see: . To enable this feature, simply add the --tool-call-parser qwen3_coder flag when launching the server:
python -m sglang.launch_server --model-path /path/to/your/model --tool-call-parser qwen3_coderMini Program Development
Nex-N1 is optimized for mini program development. For optimal performance, we recommend using Claude Code configured with both context7 and a search MCP.
claude mcp add --transport http context7 https://mcp.context7.com/mcp --header "CONTEXT7_API_KEY: [CONTEXT7_API_KEY]"
claude mcp add --transport stdio serper-search --env SERPER_API_KEY=[SERPER_API_KEY] -- npx -y serper-search-scrape-mcp-server
Refer to for more details on setting up context7.
Features & Capabilities
| Mode | chat |
| Context Window | 131,072 tokens |
| Max Output | 163,840 tokens |
| Function Calling | Supported |
| Vision | Not supported |
| Reasoning | Not supported |
| Web Search | Not supported |
| Url Context | Not supported |
Technical Details
| Architecture | DeepseekV3ForCausalLM |
| Model Type | deepseek_v3 |
| Library | transformers |
API Usage
from openai import OpenAI
client = OpenAI(
base_url="https://api.haimaker.ai/v1",
api_key="YOUR_API_KEY",
)
response = client.chat.completions.create(
model="nex-agi/deepseek-v3.1-nex-n1",
messages=[
{"role": "user", "content": "Hello, how are you?"}
],
)
print(response.choices[0].message.content)Frequently Asked Questions
What is the context window of DeepSeek V3.1 Nex N1?
DeepSeek V3.1 Nex N1 (nex-agi/deepseek-v3.1-nex-n1) has a 131,072-token context window and supports up to 163,840 output tokens per request.
How much does DeepSeek V3.1 Nex N1 cost?
DeepSeek V3.1 Nex N1 is priced at $0.14 per 1M input tokens and $0.50 per 1M output tokens when accessed via the haimaker.ai OpenAI-compatible API.
What features does DeepSeek V3.1 Nex N1 support?
DeepSeek V3.1 Nex N1 supports function calling.
How do I use DeepSeek V3.1 Nex N1 via API?
Send requests to https://api.haimaker.ai/v1/chat/completions with model "nex-agi/deepseek-v3.1-nex-n1" using any OpenAI-compatible SDK. Authentication uses a Bearer API key from https://app.haimaker.ai.
Use DeepSeek V3.1 Nex N1 with the haimaker API
OpenAI-compatible endpoint. Start building in minutes.