Technical Specifications
Model Architectures
Compare our suite of foundation models optimized for various use cases, from real-time inference to complex reasoning tasks.
| Model | Parameters | Context Window | Modalities | Best For |
|---|---|---|---|---|
| Nexus-V3-Ultra | 175B | 128K | Text, Vision, Audio | Complex reasoning, multi-modal analysis |
| Nexus-V3-Pro | 70B | 128K | Text, Code | Enterprise applications, code generation |
| Nexus-V3-Lite | 13B | 32K | Text | Edge deployment, low-latency tasks |
| Nexus-Embed-3 | 3.2B | 8K | Text | Semantic search, vector embeddings |
Quantization Support: All models support 4-bit and 8-bit quantization for edge deployment via GGUF and ONNX formats. Quantized models maintain >95% accuracy with 4x speedup.
Framework Compatibility
Python SDK
- Minimum Version3.9+
- Async SupportNative
- TensorFlow2.15+
- PyTorch2.2+
Node.js SDK
- Minimum Version18 LTS+
- Deno1.35+
- Bun1.0+
- TypeScript5.1+
Go SDK
- Minimum Version1.21+
- ConcurrencyGoroutines
- Protobufv3
- gRPCNative
API Reference
RESTful API endpoints with WebSocket support for streaming responses. All requests require Bearer token authentication.
Python
import nexus # Initialize client client = nexus.Client( api_key="nx_live_...", region="us-east-1", timeout=60 ) # Chat completion response = client.chat.create( model="nexus-v3-pro", messages=[ {"role": "user", "content": "Analyze this data"} ], temperature=0.7, max_tokens=2048, stream=True ) for chunk in response: print(chunk.choices[0].delta.content, end="")
cURL
curl https://api.nexusai.dev/v1/chat/completions \\ -H "Authorization: Bearer nx_live_..." \\ -H "Content-Type: application/json" \\ -d '{ "model": "nexus-v3-pro", "messages": [ {"role": "system", "content": "You are a helpful assistant"}, {"role": "user", "content": "What is quantum computing?"} ], "temperature": 0.7, "top_p": 0.9, "max_tokens": 1024, "stream": false }'
Core Endpoints
| Method | Endpoint | Description |
|---|---|---|
| POST | /v1/chat/completions | Generate chat completions with streaming support |
| POST | /v1/embeddings | Create vector embeddings for text inputs |
| POST | /v1/images/generations | Generate images from text prompts |
| POST | /v1/audio/transcriptions | Transcribe audio files to text (WAV, MP3, M4A) |
| GET | /v1/models | List available models and their metadata |
Performance Benchmarks
Measured on AWS p4d.24xlarge instances (8x NVIDIA A100 80GB) with optimized inference engine.
Latency
- Time to First Token (TTFT)< 50ms
- Token Generation120 tok/s
- P99 Latency120ms
- P95 Latency85ms
Throughput
- Requests/sec (Pro)2,400
- Requests/sec (Ultra)800
- Batch SizeUp to 256
- Max Concurrency10,000
Accuracy
- MMLU89.2%
- HumanEval84.5%
- GSM8K92.1%
- TruthfulQA78.3%
Performance Variance: Actual performance may vary based on input length, temperature settings, and system load. Dedicated instances guarantee consistent latency within SLA bounds.
Deployment Options
Deploy NexusAI models in the cloud, on-premise, or at the edge with our flexible infrastructure options.
Cloud API
- Regions12 Global
- Availability99.99% SLA
- Auto-scalingYes
- Managed UpdatesRolling
On-Premise
- DockerSupported
- KubernetesHelm Charts
- GPU RequiredYes
- Data SovereigntyFull Control
Edge Runtime
- Binary Size45MB
- CPU InferenceNexus-V3-Lite
- MobileiOS/Android
- Offline ModeSupported
System Requirements
| Component | Minimum | Recommended |
|---|---|---|
| CPU | 8 Core (AVX2) | 16 Core (AVX-512) |
| RAM | 32 GB | 64 GB |
| GPU | NVIDIA A10G (24GB) | NVIDIA A100 (80GB) |
| Storage | 200 GB NVMe | 1 TB NVMe |
| Network | 1 Gbps | 10 Gbps |
| OS | Ubuntu 22.04 LTS | Ubuntu 24.04 LTS |
Docker Compose
# docker-compose.yml version: '3.8' services: nexus-gateway: image: nexusai/gateway:latest ports: - "8080:8080" environment: - NEXUS_API_KEY=${NEXUS_API_KEY} - NEXUS_MODEL=nexus-v3-pro deploy: resources: reservations: devices: - driver: nvidia count: 1 capabilities: [gpu,compute]
Security & Compliance
Enterprise-grade security with comprehensive compliance certifications and data protection measures.
Encryption
- Data in TransitTLS 1.3
- Data at RestAES-256-GCM
- Key ManagementAWS KMS / HashiCorp
- PII DetectionBuilt-in Filter
Certifications
- SOC 2 Type IICertified
- ISO 27001Certified
- GDPRCompliant
- HIPAABA Available
Access Control
- AuthenticationOAuth 2.0 / SAML
- RBACFine-grained
- Audit LogsImmutable
- IP WhitelistingSupported
Data Retention: By default, input data is not stored after inference. Enterprise plans can enable audit logging with configurable retention periods (7, 30, or 90 days). All data is processed in isolated tenant environments.
Rate Limits & Quotas
API rate limits by tier. Limits can be increased upon request for enterprise customers.
| Limit | Free Tier | Pro | Enterprise |
|---|---|---|---|
| Requests / Minute | 60 | 1,000 | Custom |
| Requests / Month | 100K | 5M | Unlimited |
| Max Tokens / Request | 4,096 | 32,768 | 128,000 |
| Max Tokens / Minute | 50K | 500K | Custom |
| Parallel Requests | 3 | 20 | 100+ |
| Context Window | 32K | 128K | 128K |
Response Codes
| Code | Meaning | Action |
|---|---|---|
| 429 | Rate Limit Exceeded | Implement exponential backoff. Check Retry-After header. |
| 400 | Bad Request | Check payload format and model parameters. |
| 401 | Unauthorized | Verify API key validity and permissions. |
| 500 | Internal Server Error | Transient issue. Retry with delay. Contact support if persistent. |