Company profile · data as of 2026-10-15
BenchFlow
Builds stateful environments, benchmarks, and runtime infrastructure for training capable AI agents.
Open the full interactive profile
- Website
- benchflow.ai
- Primary offering
- Frontier data & RL environments
- Category
- Agent environments and evals
- Tags
- Agent environments and evals, Open source, Workplace simulation, Procedural skills, Robotics, Post-training
What BenchFlow sells
Builds practice tasks and tests for AI programs that carry out work. It also provides tools to run those tests.
- SkillsBench — Tests whether reusable instructions improve agents’ task performance.
- ClawsBench — Replicas of Gmail and Calendar, plus Docs, Drive and Slack, that retain changes between actions. Scores capability and safety separately.
- Robo Use — Lets coding agents operate simulated robots through shell commands.
- PostTrain Arena — Community environments tested using a fixed training recipe. Training runs are currently paused.
- BenchFlow runtime — Open-source tools for running and scoring agents in test environments.
Headline metrics
| Metric | Value | As of | Source |
|---|---|---|---|
| Total funding (reported) Total Funding | at least $1M | 2026-10-12 | Build the future of AI evals; funding and hiring announcement |
| Other metric SkillsBench GitHub stars | at least 1,800 | 2026-10-14 | BenchFlow: /about |
Most recent figure per metric kind. Run rate, ARR, revenue, funding and valuation are kept distinct; estimates and forecasts are marked.
Paid work and open routes
- Infrastructure internships — Not published (historical)
- PostTrain environment contributions — Paid route not established (paused)
Pay labels are quoted as published and may be out of date; check each listing.
Explore further
Open the full interactive profile
The interactive profile shows every sourced claim, funding round, relationship and work listing with its evidence grade. Browse all companies.