Prime Intellect launched verifiers 0.2.0, previewing a rewritten core shipped under the new verifiers.v1 namespace. The v1 rebuilds environments to run agentic workloads at scale by breaking the previously bundled environment into three composable pieces: a taskset (data, tools, scoring), a harness (solves the task, produces a rollout), and a runtime (local or sandbox). The architecture centers on a verifiers-managed interception server that proxies requests between the agent's runtime and the inference server, recording traces and rewriting tool responses to mitigate reward hacks. Each server multiplexes a constant number of rollouts, defaulting to 32, with a pool scaling elastically. Verifiers supports three dialects: OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. Compared to v0, v1 features trace growth linear in turns (unique nodes) instead of quadratic, supports non-linear rollouts via branches, and allows any compatible harness (e.g., Codex, Terminus 2). Prime Intellect ported Terminal Bench 2 into v1 with only a small class, and in internal testing, verifiers matched Harbor’s performance on the same tasks.