BenchGen
BenchGen is the learning infrastructure for AI agents: an open platform where developers discover benchmarks and RL environments, evaluate their complete agent system — model and harness together — against verifiable rewards, and export clean trajectory data for fine-tuning. One loop: benchmark → evaluate → fine-tune → re-evaluate.
Learn more
Inkling
Inkling is an open-weights multimodal AI model from Thinking Machines designed as a customizable foundation model for developers, researchers, and enterprises. The model is a Mixture-of-Experts transformer with 975 billion total parameters, 41 billion active parameters, and support for context windows up to 1 million tokens. Inkling was trained from scratch on text, images, audio, and video, giving it native capabilities across reasoning, coding, agentic tool use, vision, audio, factuality, and instruction following. It is built with controllable thinking effort so users can balance performance, latency, and token efficiency for different workloads. The model is available for fine-tuning on Tinker, with playground access, API availability through ecosystem partners, and full weights published on Hugging Face. Built for customization, Inkling gives teams an open-weights base model for building domain-specific AI systems, multimodal agents, coding workflows, research tools, and more.
Learn more
Overmind
Overmind is a model training and deployment platform for teams building AI agents. It ingests production traces through OpenTelemetry and creates datasets from trace data, file uploads, and connectors for Langfuse, LangSmith, Braintrust, and Galileo. Teams can inspect agent context and runtime behavior, prepare and score datasets, evaluate prompts and models against their own examples, and optimize agent behavior through scored changes. The platform supports LoRA and full fine-tuning of open-weight models, benchmarking with configured evaluators, and deployment through an OpenAI-compatible endpoint; trained weights can also be downloaded. Users can work in the hosted console or run the platform on their own infrastructure with Docker Compose. It is designed for AI engineering and product teams operating agents in production that want to test, improve, or specialize models using their own data.
Learn more
Tuning Engines
Tuning Engines is a unified AI control and governance layer for teams building production intelligence across models, agents, tools, and fine-tuned systems.
It brings together the full AI lifecycle in one governed platform: inference, model routing, fallback policies, fine-tuning jobs, datasets, evaluations, model imports and exports, custom models, agents, MCP servers, reusable skills, guardrails, AGT YAML policies, data capture, runtime traces, usage analytics, API keys, billing, team roles, and integrations.
Developers get OpenAI-compatible APIs, Anthropic-compatible routes, CLI workflows, MCP access, coding-agent integrations, and resource catalogs for models, agents, tools, and skills. Teams can connect Claude Code, OpenCode, Aider, Cline, Roo, Continue.dev, Cursor, VS Code, Windsurf, and other AI workflows through a single governed platform.
Learn more
