baseten

Portal hopping with Stable Diffusion 👾

Open, independent benchmark on LLM inference providers: same GLM 5.3 Flash, 600 paired requests each. Latency, tokens per second and task success. Baseten, DeepInfra, Fireworks AI, Modal, Nebius, Novita AI, Parasail, Telnyx, Together AI, Z.AI. Python runner and public data.

Mantis - unified perception-reasoning-action agent for computer use

The open control plane for multi-model AI

Kosmos: VLMs for world-model robot-policy evaluation.

One-shot GPU harness for traffic-grounded teacher, student, and route evaluation

Fail-safe, low-latency LLM serving: a dependency-free simulator + Baseten design study of KV-cache saturation, metastable collapse, and graceful degradation.

Agentic Investment research copilot

LLM ingestion pipeline processing schematic electrical diagrams into instructions and BOMs made with Python and Baseten's Kimi K3 Vision LLM model. Features a DJI RoboMaster EP autonomous vehicle programmed with a Raspberry Pi 5 to follow the LLM instructions via April Tag sensors and retrieve materials. Submitted for Hack the North 2026.

AI Lab — ML/LLM Engineering Platform. Unified provider registry, model catalog, experiment tracking, cost analytics, and OpenCode-native config sync. 10 providers, REST API, CLI, Web UI.

C# SDK for the Baseten API -- model serving and deployment platform with Truss framework