Proliferate routes all LLM requests through its built-in proxy. You never need to bring your own API keys — every supported model works out of the box. On Proliferate Cloud, the latest models are available within 20 minutes of release.
Self-hosted deployments work with any model provider. Bring your own API keys and route through Azure OpenAI, AWS Bedrock, Google Vertex AI, or connect directly to any provider. See the self-hosting guide and environment configuration for setup.
Supported Models
| Model | Provider | Best For |
|---|
| Claude Sonnet 4.6 (default) | Anthropic | Agents and coding tasks. The best balance of speed, cost, and quality. |
| Claude Opus 4.6 | Anthropic | Complex multi-step tasks requiring deep reasoning. |
| GPT-5.2 | OpenAI | OpenAI flagship thinking model. |
| Gemini 3 Pro | Google | Large-context tasks (1M token context window). |
| Gemini 3 Flash | Google | Fast, lightweight tasks where speed matters most. |
| DeepSeek V3 | DeepSeek | Strong open-weight coding model. |
| DeepSeek R1 | DeepSeek | Reasoning-heavy tasks with chain-of-thought. |
| Grok 4.1 | xAI | xAI flagship model. |
| Grok 4.1 Fast | xAI | Quick xAI tasks where latency is critical. |
| Codestral | Mistral | Code-specialized tasks from Mistral. |
Claude Sonnet 4.6 is the default model for all new sessions and automations. It offers the best combination of coding ability, speed, and cost efficiency.
Providers
Proliferate supports models from six providers:
- Anthropic — Claude Sonnet 4.6, Claude Opus 4.6
- OpenAI — GPT-5.2
- Google — Gemini 3 Pro, Gemini 3 Flash
- DeepSeek — DeepSeek V3, DeepSeek R1
- xAI — Grok 4.1, Grok 4.1 Fast
- Mistral — Codestral
No API Keys Required
Proliferate’s LLM proxy handles all authentication between the agent and the model provider. This means:
- No key management — You never create, rotate, or store API keys for any model provider.
- Sandboxes never see real keys — The proxy generates scoped, short-lived virtual keys for each session. The sandbox environment has no access to your actual provider credentials.
- Per-org spend tracking — All model usage is tracked and attributed to your organization automatically. You can monitor spend across models and sessions from the dashboard.
Self-hosted deployments bring their own API keys and can route through Azure OpenAI, AWS Bedrock, Google Vertex AI, or any provider directly. See environment configuration for details. Reasoning Effort
Some models support configurable reasoning effort, which controls how much “thinking” the model does before responding. This lets you trade speed for depth depending on the task.
| Level | Behavior |
|---|
| Quick | Minimal reasoning. Fastest responses. Best for simple, well-defined tasks. |
| Normal (default) | Balanced reasoning. Good for most coding tasks. |
| Deep | Maximum reasoning depth. Best for complex architecture decisions, debugging, and multi-step planning. |
- Claude Sonnet 4.6
- Claude Opus 4.6
- GPT-5.2
- DeepSeek R1
- Grok 4.1
Models without reasoning support (Gemini 3 Pro, Gemini 3 Flash, DeepSeek V3, Grok 4.1 Fast, Codestral) ignore the reasoning effort setting and always use their default behavior.
Choosing the Right Model
| Use Case | Recommended Model | Why |
|---|
| General coding tasks | Claude Sonnet 4.6 | Best all-around coding agent. Fast, accurate, cost-effective. |
| Complex refactoring | Claude Opus 4.6 | Deeper reasoning handles large-scale architectural changes. |
| Large codebase analysis | Gemini 3 Pro | 1M context window fits entire codebases in a single pass. |
| Quick fixes and typos | Gemini 3 Flash | Fastest response time for simple changes. |
| Debugging with chain-of-thought | DeepSeek R1 | Explicit reasoning traces help with hard-to-find bugs. |
| Cost-sensitive workloads | DeepSeek V3 | Strong open-weight model at lower cost. |
| Code generation tasks | Codestral | Purpose-built for code by Mistral. |
Configuring Models
Per session
When starting a session from the web dashboard or terminal, select the model from the model picker before sending your first message. You can also change the model mid-session.
Per automation
Each automation has a model setting in its configuration. Set it when creating or editing the automation:
- Go to the automation settings
- Under Model, select the model and optionally set reasoning effort
- Save the automation
All runs triggered by that automation will use the configured model.
Via the API
When creating sessions or automations through the API, pass the modelId field:
{
"modelId": "claude-sonnet-4.6",
"reasoningEffort": "normal"
}
claude-sonnet-4.6, claude-opus-4.6, gpt-5.2, gemini-3-pro, gemini-3-flash, deepseek-v3, deepseek-r1, grok-4.1, grok-4.1-fast, codestral. 