LLM Integrations
Connect your Paperclip agents to any major LLM provider. Switch models without redeploying.
OpenAI
Broadest ecosystem, ultra-cheap high-volume tier, and million-token context at every price point.
Anthropic
Exceptional instruction following with extended thinking and 1M-token context from Sonnet 5 up to the new Fable 5.1 flagship.
Google Gemini
Largest context window we host, with native multimodal ingest and Search grounding.
Mistral AI
Flagship at $0.50/1M input, open-weight options, and EU data processing.
Which LLM should your agent run on?
Start with Claude Sonnet 5 or GPT-5.6 Terra — both cost $2 per 1M input tokens and handle most production agent workloads. Route high-volume simple tasks to GPT-5.6 Luna ($0.20) or Mistral Small 3.1 ($0.10), long-document and multimodal work to Gemini 3.1 Pro (2M-token context), genuinely hard reasoning to Opus 5 or GPT-5.6 Sol, and the most demanding long-horizon evaluations to Anthropic's newest flagship, Claude Fable 5.1 ($10/$50, 1M context). On Paperclip you can mix all four providers per agent and switch at any time without redeploying.
How It Works
Add your API key
Save your LLM provider key as an environment variable. Encrypted at rest.
Choose a model
Select from the dropdown — GPT-5.6, Claude Sonnet 5, Gemini 3.1 Pro, or any supported model.
Deploy
Your agent is live. Switch models at any time without redeploying.
Model Comparison (September 2026)
| Provider | Flagship | Context | Cheapest tier | Best For |
|---|---|---|---|---|
| OpenAI | GPT-5.6 Sol ($5/$30) | 1.05M | Luna $0.20/$1.20 | General use, tool calling |
| Anthropic | Claude Fable 5.1 ($10/$50) | 1M | Sonnet 5 $2/$10 | Instruction following, analysis |
| Gemini | Gemini 3.1 Pro ($2/$12) | 2M | Flash-Lite $0.30/$2.50 | Long context, multimodal |
| Mistral | Mistral Large 3 ($0.50/$1.50) | 262K | Small 3.1 $0.10/$0.30 | Cost efficiency, multilingual |
Lineups and prices verified against OpenAI's public API pricing, Anthropic's public pricing documentation, Google's Gemini API pricing, and Mistral's model documentation in September 2026.
Integrations FAQ
Doesn't OpenAI, Anthropic, and Google already host agents?
Each vendor now ships some form of managed agent execution — OpenAI's Agents API, Anthropic's Managed Agents, Google's Vertex AI Agent Builder — but each locks you to its own models and stack. Paperclip is the provider-neutral runtime: your agent's process supervision, persistent memory, restarts, and monitoring work identically across all four providers, and you can route each run to whichever model fits.
Can I use multiple LLM providers on the same agent?
Yes. Model routing sends each run to the provider that fits: high-volume tasks to GPT-5.6 Luna or Mistral Small, long documents to Gemini, hard reasoning to Claude or GPT-5.6 Sol. One hosted bill instead of four API invoices.
Do I need separate API keys for each provider?
Yes — each provider bills its own key. Add them as encrypted environment variables in the HostAgentes dashboard, and switch or route between them at any time.
How much do the LLM integrations cost?
Integrations are included on every Paperclip plan (from $15/mo); you pay the providers directly for token usage. Current flagships: Claude Fable 5.1 $10/1M input, GPT-5.6 Sol $5/1M input, Claude Opus 5 $5/1M, Gemini 3.1 Pro $2/1M, Mistral Large 3 $0.50/1M — with cheaper small-model tiers from $0.10/1M (verified September 2026).