Today, we’re launching SmartInference—a revolutionary addition to the Raindrop platform that brings 30+ state-of-the-art AI models directly to your fingertips. For the first time, developers can access models like Kimi K2’s 1 trillion parameters and GPT-OSS 120B through a single multi-agentic AI platform, with performance that rivals enterprise deployments: Example: 250+ tokens per second throughput at just 350ms latency for GPT-OSS-120B
Most agentic platform vendors only support models up to 70B models or completely leave the model selection, routing, and operation up to you. Raindrop SmartInference takes a different approach. You describe the application requirement, and SmartInference chooses the best model to meet your needs (obviously, human model selection is available too).
Every developer building AI agents faces the same frustrating reality: you’re either limited to smaller models that can’t handle complex reasoning, or you’re managing a corn maze of API keys, rate limits, and infrastructure decisions. Want to use DeepSeek R1’s 671B parameters for advanced reasoning? That typically requires an enterprise contract. Need Whisper Large v3 for transcription alongside Kimi K2 for complex tool use? Good luck orchestrating multiple providers.
Even worse, most platforms force you to become an AI infrastructure expert just to get started. You need to understand model capabilities, optimize for latency, manage costs, and handle failovers—all before writing a single line of your actual application.
SmartInference eliminates these barriers completely. Through our SmartInference process, Raindrop automatically picks the perfect model for each task in your application. Building a customer support agent? It might select Llama 3.1-8b-instant for greeting and routing (ultra-fast responses), then seamlessly leverage DeepSeek R1 for complex problem-solving that requires deep reasoning.
You focus on what your agent should do. SmartInference handles everything else.
Kimi Family
GPT-OSS Family
DeepSeek Family
Llama Family
These aren’t experimental models running on underpowered infrastructure. Our SmartInference service delivers:
This is the same performance enterprises expect from their private deployments—now available to every developer for just $5/month during our public beta.
Here’s how simple it is to create a sophisticated multi-model agent with Raindrop and SmartInference:
raindrop build deploy
SmartInference unlocks a whole new class of applications with bigger, smarter models that are only used when needed. Build agents that can:
SmartInference is available immediately to all Raindrop Users.
You get:
You can build production agents in 25 minutes with zero infrastructure or backend skills.
That’s not marketing speak. That’s the reality of what our users are already doing. They’re building applications that are “more than just a database”, things that would typically require a team of ML engineers and a significant cloud provider budget and time. They’re deploying in minutes, not months. They’re iterating rapidly without worrying about infrastructure.
The age of being limited by model availability is over. Sign up for Raindrop public beta access at liquidmetal.run and start building with SmartInference today.
For $5/month during public beta, you can build any agent, MCP, API, or production backend you can conceive—powered by the most advanced AI models on the planet.
Don’t wait. This pricing will never be offered again after beta. Early adopters get grandfathered for life.
Raindrop SmartInference is available now as part of the Raindrop public beta. Setup takes less than 5 minutes. Visit liquidmetal.run to get started.