Aster is a Public Benefit Corporation that develops autonomous research systems to compress decades of scientific discovery into hours. The company applies its technology to build a high-performance inference API, currently offering the fastest gpt-oss-120b deployment on GPU at 644 tokens per second. Its platform coordinates populations of research agents that continuously optimize the underlying stack, with zero data retention by default.
Funding
Funding not disclosed
Founders
Product
Problem
Discovering novel technologies and conducting scientific research traditionally takes years, requiring extensive manual experimentation, literature review, and iterative testing. This slow pace limits the rate of innovation and delays the application of new discoveries across industries.
Solution
Aster develops autonomous research systems that compress decades of research into hours by coordinating populations of AI agents that design, run, and analyze experiments at massive scale. The company has demonstrated results across benchmarks including NanoChat, ProteinGym, and the NanoGPT speedrun, and has scaled its systems to process millions of tokens per second. Aster also commercializes its research infrastructure as a high-performance inference API, offering access to leading models like gpt-oss-120b, Kimi K3, and GLM 5.2 with optimized speed and cost. The system continuously improves itself by applying autonomous research to optimize the very stack it runs on, compounding efficiency gains over time.
Target Audience
Primary customers are AI developers, researchers, and enterprises that need high-throughput, low-latency inference for production workloads, as well as scientific organizations seeking to accelerate research and experimentation through autonomous systems.
Features
- Autonomous research platform that coordinates thousands of agents for parallel experimentation and analysis
- Inference API delivering the fastest gpt-oss-120b deployment on GPU at 644 tokens per second, outperforming major providers
- Access to multiple frontier models including gpt-oss-120b-fast, Kimi K3 (2.8T-parameter MoE with 1M context), and GLM 5.2
- Zero data retention by default, with prompts and outputs processed in memory and only token counts stored for billing
- OpenAI-compatible API supporting standard SDKs and tools including LangChain, LiteLLM, Cursor, and the Vercel AI SDK
- Self-optimizing infrastructure where research agents continuously improve inference performance and cost efficiency
- Tiered pricing plans from free to enterprise with custom token volumes and dedicated endpoints