github PrimeIntellect-ai/verifiers v0.1.12.dev0

8 hours ago

Verifiers v0.1.12.dev0 Release Notes

Date: 03/22/2026

Highlights since v0.1.11

  • Added the new opencode_rlm_env plus broader opencode environment cleanups and performance work, including executor autoscaling and incremental metrics for better throughput under load.
  • Improved runtime reliability with more robust cancellation handling, env server startup tuning, safer port allocation, and better event loop lag monitoring.
  • Expanded configuration and lifecycle ergonomics with Rubric cleanup/teardown hooks, configurable eval output_dir, preserved multimodal media in saved eval results, and ClientConfig.extra_headers_from_state.
  • Refined rubric and environment performance with lazy imports, faster file I/O, math-rubric optimizations, and follow-up fixes to logging and response normalization.
  • Added fresh docs for BrowserEnv integration and a new environment performance guide.

Changes included in v0.1.12.dev0 (since v0.1.11)

Environments, execution, and performance

  • deprecate RolloutGatewayMixin (#1017)
  • Lazily import packages (#1019)
  • Add BrowserEnv integration README (#1020)
  • tune GC on env server before accepting requests (#1022)
  • opencode_rlm_env (#1023)
  • Preserve multimodal media in saved eval results (#1015)
  • Add cleanup and teardown lifecycle hooks to Rubric (#1026)
  • remove redundant msg normalization + align env_response api (#1027)
  • make output_dir configurable in evals (#1029)
  • misc improvements to opencode envs (#999)
  • perf improvs for opencode envs + math rubric (#1034)
  • fix: task cancelation race + RLM sandbox workers (#1035)
  • perf: incremental metrics (#1036)
  • perf: offload file i/o to thread pool (#1037)
  • feat: improve event loop lag monitor (#1038)
  • perf: executor autoscaling (#1039)

Reliability, cancellation, and bug fixes

  • Fix get_free_port_pair() TOCTOU race condition (#1013)
  • fix display of custom sampling args (#1025)
  • fix output dir logging (#1041)
  • fix: revert opencode_env config regression and move RLM logic out of cli_agent_env (#1042)
  • chore: reuse math rubric in hybrid math rubric (#1043)
  • fix cancelled + serialize error (#1044)
  • docs: add performance guide for environments (#1045)
  • perf: math rubric skip overlong answers (#1046)
  • fix: call uncancel() after catching CancelledError in process_request (#1047)
  • feat: add extra_headers_from_state to ClientConfig (#1048)

Full Changelog: v0.1.11...v0.1.12.dev0

Don't miss a new verifiers release

NewReleases is sending notifications on new releases.