Bespoke Labs

What Bespoke Labs does

Bespoke Labs is an applied AI research lab that builds environment infrastructure for the agent revolution. They create company-scale RL environments and optimization systems to help frontier labs and enterprises train long-horizon agents for production. Their work focuses on enabling agents to master complex workflows and ensuring their reliability in real-world scenarios. They also conduct open research, develop benchmarks, and provide deployed optimization infrastructure.

Products

• OpenThoughtsOne of the best open reasoning datasets, with over 10k monthly downloads on Hugging Face.
• Terminal-BenchThe first environment-based benchmark for agentic systems, cited by Anthropic, OpenAI, and Google DeepMind.
• GEPAAn optimizer that uses evolutionary algorithms for prompt optimization, used by over 200 teams in production.

Key capabilities

• Company-scale simulations with real codebases and microservices
• Automated prompt and policy searches via GEPA optimizer
• Production-grade RL and benchmarking
• Collaborative research in RL, curation, and benchmarking
• Environments that mirror systems and processes for agent evaluation
• Optimization of prompts and policies automatically with GEPA and RL

Use cases

• Training long-horizon agents for production
• Teaching new tasks to AI agents
• Improving AI agent reliability
• Evaluating AI agents in production-like environments
• Optimizing prompts and policies for AI agents

AI approach

Bespoke Labs is an applied AI research lab that builds company-scale RL environments and infrastructure to train long-horizon agents for production. They focus on creating realistic, complex environments for agents to practice and improve, and develop optimization systems like GEPA for automated prompt and policy searches. They also contribute to open research, benchmarks, and deployed optimization infrastructure.

Tech named: RL environments, GEPA optimizer, Evolutionary algorithms for prompt optimization, OpenThoughts (open reasoning dataset), Terminal-Bench (environment-based benchmark for agentic systems)

What it says sets it apart

• Focus on environment research and infrastructure for the agent revolution
• Building company-scale RL environments with real codebases and microservices
• Proprietary GEPA optimizer for automated prompt and policy searches
• Strong research track record with open research, benchmarks, and deployed optimization infrastructure
• Cited by major AI research labs like Anthropic, OpenAI, and Google DeepMind
• Team built by researchers with experience from frontier labs, Big Tech, and top academic institutions (ex-Google DeepMind, UC Berkeley professor)

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.