Replit Ships Model Selector So Developers Stop Overpaying for Simple Fixes

Replit just shipped Model Selector, a new feature that lets you choose which AI model powers your Agent session and dial in exactly how much reasoning effort it applies. The headline addition: open-weight models are now first-class citizens alongside Claude, GPT, and Gemini. For anyone who has ever wished they could swap in a cheaper model for a quick fix and a heavier one for a hard architectural problem, this is the feature you've been waiting for.

One knob for intelligence, one for effort

Model Selector has two moving parts. The first is straightforward model choice: you can now pick from a growing catalog that includes both closed frontier models and open-weight alternatives. The second is Effort Control, which maps directly onto Replit's existing Agent mode system.

Agent modes let you control the balance between speed, cost, and capability when using Agent. The current tiers are:

• Lite , optimized for quick edits, using fast, lightweight models for visual tweaks, bug fixes, and other small, scoped changes.
• Economy , best default when you want strong results without burning credits on the most capable models.
• Power , for the most capable standard builds; turn on Turbo when you need the fastest runs.
• High Effort (toggle) , when enabled, Agent performs deeper, more deliberate reasoning and invokes its most capable frontier models, improving outcomes on the most complex tasks.
• Turbo (toggle) , gives 2.5x faster responses using the fastest models, but requests cost about 2x more than Power.

The key insight is that High Effort is selective, not blanket. Agent applies that extra power selectively, routing to the more capable models only when a request is genuinely hard, not on every run. That means leaving High Effort on during a complex refactor adds cost only on the hard parts, not on every line it touches.

Why open-weight models matter here

Open-weight models (models whose trained parameters are publicly released, letting anyone run or fine-tune them) have closed the gap with closed frontier systems dramatically over the past year. The best open-weights AI models now rival the closed systems from OpenAI and Anthropic on most everyday business tasks. The cost difference, however, remains enormous.

DeepSeek V4 Flash is the first open-weight model that teams immediately dropped into real agentic pipelines as a plausible substitute for an Anthropic- or OpenAI-class frontier model. The larger V4 Pro variant set the ceiling with a score of 80.6% on SWE-bench Verified , the top open-weights score. But it is Flash that broke through, because it captures most of that capability at a price that is on the pareto frontier of performance and cost. That is roughly 150x cheaper than GPT-5.5's output costs.

The open-weight ecosystem is also moving fast. Between July 1 and July 22 alone, ten new model families reached OpenRouter's catalog , roughly one every 2.2 days , spanning OpenAI's renamed GPT-5.6 trio, xAI's Grok 4.5, Moonshot's Kimi K3, Meta's first paid-API model, Thinking Machines' debut, two coding specialists from Kuaishou, Meituan's LongCat 2.0, Google's triple Gemini launch, and poolside's open-weight Laguna line. Keeping up with that pace manually is unrealistic, which is exactly why Replit added recommended defaults to the selector.

The open-weight access layer

Under the hood, Replit's open-weight access routes through OpenRouter. OpenRouter is a pass-through gateway to 200+ models from providers such as Meta, Mistral, DeepSeek, Qwen, Microsoft, Nvidia, and Amazon , browse the live list at openrouter.ai/models. Replit manages the credentials and privacy settings so you don't have to wire anything up yourself.

Privacy defaults are worth knowing: for individual and team users, Replit configures OpenRouter requests with paid endpoints training disabled, meaning your data will not be used for training by paid model providers. Free-tier model endpoints carry different terms, so check before routing sensitive code through them.

The bigger picture: Replit signs the open-weights letter

This feature didn't ship in a vacuum. Replit was among the companies that signed the open-weights letter alongside Nvidia, Mistral, Mozilla, Palantir, Perplexity, ServiceNow, Telnyx, and Y Combinator. The letter's argument is straightforward: open weights let every organization match the right model to the right job at the right cost, reserving frontier-scale capability for genuine frontier problems , the discipline that will make AI economically sustainable as its use scales into billions of everyday tasks.

That philosophy is now baked directly into the product. Recommended defaults help users who aren't tracking the latest releases, while power users can override and pick exactly what they want. The combination of model choice plus effort control means you're no longer forced to choose between cost and quality on a per-session basis , you can make that call per task.

Who gets it and what it costs

Model Selector is available now for all Core and Pro users. Open any project, tap the model selector in the Agent chat input, and make your choice.

• Core is $20/month and Pro is $100/month, with a free Starter tier and custom Enterprise pricing.
• Core includes $25 of monthly usage credits, up to 5 collaborators, and parallel work with up to 2 agents.
• Pro targets bigger teams, bundling $100 of monthly credits, up to 15 builders, up to 10 parallel agents, and the most powerful models.
• Turbo mode (the fastest, most expensive option) is available on Pro and Enterprise only.

Credits are consumed by model calls, compute, and deployments. Once you use up your monthly credits, Replit switches to pay-as-you-go billing for the overage. If you plan to run Power + High Effort on complex builds regularly, budget accordingly , or use Economy for the bulk of your work and reserve the heavy modes for genuinely hard problems. That's the whole point of having the dial.