AgentdexBlog · · Agentdex Editorial
What LLM Does Replit Use? (2026)
Replit (Agentdex profile) doesn't use a single LLM. Replit Agent (Agentdex profile) routes each request to models from several labs, including OpenAI and Anthropic, and the model you get depends on your Agent mode and plan. On paid plans you can also pick a model yourself. This guide explains how Replit chooses models as of October 24, 2026, which ones are available, and how model choice affects cost.
Key Takeaways
- By default, Replit Agent uses intelligent model routing: it picks a model for each task, balancing quality and cost.
- Agent has three modes. Free Mode always routes automatically. Power Mode and Max Mode use monthly credits, and on Core and Pro you can choose the model manually.
- Recent options include OpenAI's GPT-6 family, Anthropic's Claude Sonnet and Opus models, and Kimi K3. The list changes as models roll out.
- For AI features inside your own app, Replit AI Integrations give managed access to OpenAI, Anthropic, Google Gemini, Grok and OpenRouter models, billed to your credits at public API prices.
The short answer
Replit Agent is a system built on top of third-party frontier models, not one model of its own. Replit's docs describe "intelligent model routing": for each request, Replit picks the model it judges best for the task at a sensible cost. That is why you won't find one model name on Replit's homepage, and why the answer changes as new models are released.
How Agent modes choose models
Mode | How the model is chosen | Billing |
|---|---|---|
Free Mode | Always intelligent model routing; no manual choice | Included allowance (on Core and Pro it refreshes every five hours) |
Power Mode | Auto routing, or a manual pick on Core and Pro | Monthly credits |
Max Mode | Auto routing, or a manual pick from the most capable models on Core and Pro | Monthly credits |
Power Mode replaced the old Economy Mode at the same price. Max Mode is aimed at the hardest tasks and offers the most capable models.
Which models can you pick?
When you choose a model manually in Power or Max Mode, Replit's documentation currently lists:
- Power Mode: GPT-6 Sol, GPT-6 Luna Fast, Claude Sonnet 5 and Claude Sonnet 4.6.
- Max Mode: GPT-6 Astra, Claude Fable 5.1, Claude Opus 5.5, Claude Opus 4.8, Claude Opus 5, Claude Opus 5 Fast and Kimi K3.
Replit notes that the list varies by rollout, and the "Fast" variants require Pro or Enterprise. Check the model picker in your workspace for what you can use today.
You can also set an effort level, from Low through Medium, High and Extra High to Max. Higher effort lets the model think longer, and on the toughest tasks it can cost up to about twice as much. A Compare models feature (in beta) lets you send the same Chat prompt to up to four models and compare the answers.
Does Replit have its own model?
Replit has trained models before. In October 2023 it released replit-code-v1_5-3b, a small open-source code completion model under the Apache 2.0 license on Hugging Face. Today's Replit Agent relies on frontier models from other labs, routed by Replit.
Models inside your own app
Agent models are what Replit uses to build your app. If your app itself needs AI, for example a chatbot or summarizer, Replit AI Integrations give paid plans managed access to OpenAI, Anthropic, Google (Gemini), Grok and OpenRouter without setting up separate accounts. Usage is billed to your Replit credits at each provider's public API price. You can also bring your own API key, which works on the free Starter plan too.
How model choice affects cost
Replit's plans as of October 2026:
- Starter: free, with Free Mode.
- Core: $20/month ($18/month billed annually), with $20 of monthly credits and up to 30 hours of chat in Free Mode.
- Pro: $100/month ($90/month billed annually), with $100 of credits, up to 15 collaborators and 10 parallel agents.
- Enterprise: custom pricing; admins can control which model providers each workspace can use.
The practical rule: let Auto routing handle routine work, use Power Mode for normal building, and save Max Mode and high effort for hard bugs or large refactors. Picking the most expensive model for every request burns credits without always improving results. Our Replit AI guide covers the full product, and best LLM for coding compares the underlying models.
Privacy and training
Replit may use your content to improve its models. Pro users can opt out, and Enterprise content is not used for training. If you work on private client code, check this setting before you start. For other options, see Replit vs Cursor and Replit vs Bolt.
FAQ
What AI model does Replit Agent use?
It uses several. By default, intelligent model routing picks a model for each request from labs including OpenAI and Anthropic. On Core and Pro you can choose a specific model in Power or Max Mode.
Does Replit use Claude?
Yes. Replit's model list includes Anthropic's Claude Sonnet and Claude Opus models, alongside OpenAI's GPT-6 models and Kimi K3.
Can I choose the model in Replit?
Yes, on Core and Pro, in Power Mode and Max Mode. Free Mode always uses automatic routing.
Does Replit have its own LLM?
Replit released an open-source code model, replit-code-v1_5-3b, in 2023. Replit Agent today relies on third-party frontier models chosen by Replit's routing.
What happened to Replit's Economy Mode?
Power Mode replaced Economy Mode at the same price.
Written by Agentdex Editorial. Agentdex is an independent directory reviewing AI coding tools; we are not affiliated with the companies mentioned. Prices and features are as of October 24, 2026, from official sources linked above. Always confirm on the vendor's site.