Skip to main content

​
Models

Spike uses its own optimized models by default, designed specifically for real-time conversational AI with personality and memory. You can also configure your personalities to use models from other providers.

​
Default models

By default, Spike selects the optimal model for each task:
TaskDefault ModelDescription
Chatspike-chat-1Optimized for conversational responses with personality
Voice synthesisspike-voice-1Low-latency text-to-speech with emotion
Voice recognitionspike-stt-1Real-time speech-to-text
Video generationspike-avatar-1Lip-synced avatar rendering
Embeddingsspike-embed-1Memory and knowledge retrieval

​
Using other providers

You can configure personalities to use models from other providers. Spike acts as a unified interface, handling authentication, rate limiting, and failover.

​
Supported providers

Anthropic

Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku

OpenAI

GPT-4o, GPT-4 Turbo, GPT-3.5 Turbo

Google

Gemini 1.5 Pro, Gemini 1.5 Flash

​
Configuration

Set the model when creating or updating a personality: 
{
  "name": "My Assistant",
  "system_prompt": "You are a helpful assistant.",
  "model_config": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514",
    "temperature": 0.7,
    "max_tokens": 1024
  }
}

​
Fallback behavior

If your configured provider is unavailable, Spike can automatically fall back to an equivalent model: 
{
  "model_config": {
    "provider": "openai",
    "model": "gpt-4o",
    "fallback": {
      "provider": "anthropic",
      "model": "claude-sonnet-4-20250514"
    }
  }
}

​
Real-time considerations

For voice and video conversations, latency is critical. Spike’s default models are optimized for real-time use:
  • First token latency: < 200ms
  • Voice synthesis: < 150ms to first audio
  • Avatar rendering: < 300ms to first frame
When using external providers, expect slightly higher latency due to additional network hops. For latency-sensitive applications, we recommend using Spike’s default models or providers with streaming support.

​
Model selection guidelines

Use caseRecommended
Real-time voice/videoSpike default models
Complex reasoningClaude 3 Opus or GPT-4o
High throughput chatClaude 3 Haiku or GPT-3.5 Turbo
Cost optimizationSpike default or Haiku-class models
Model availability and pricing vary by plan. Contact us for enterprise model options.