Verifiers v0.1.12.dev3 Release Notes

Date: 04/06/2026

Highlights since v0.1.12.dev2

• Added max_total_tokens parameter to MultiTurnEnv for controlling total token budget.
• Introduced max_turns_in_context and summarize_turns standard tool for smarter context management in RLM.
• Hardened RLM root-tool transport by removing unsafe pickle deserialization.
• Improved RLM prompts, metrics, and scaffolding with better turn-limit awareness.
• Added support for headers/extra_headers in endpoints.toml.

Changes included in v0.1.12.dev3 (since v0.1.12.dev2)

Features and enhancements

• feat: add max_total_tokens parameter to MultiTurnEnv (#1101)
• feat: add max_turns_in_context with answer extraction and metric docs (#1099)
• feat: replace remove_conversation_turns with summarize_turns standard tool (#1095)
• feat: support headers/extra_headers in endpoints.toml (#1051)
• feat: RLM inform model about max_turns_in_context limit in scaffolding (#1111)
• feat: replace timing info in RLM REPL output with root tool time metrics (#1097)
• feat: remove token/timing info from llm_batch output and add max_turns metric (#1098)

Fixes and maintenance

• fix: opencode config for model names without slash (#1114)
• fix: composable mkdir path quoting (#1110)
• fix: tool args passing (#1106)
• fix: math rubric timeout (#1096)
• fix: pin regex<2026.4.4 for missing cp312/cp313 wheels (#1109)
• fix: bump uv requirement to >=0.11.1 (#1112)

Security and hardening

• security: harden RLM root-tool transport to remove unsafe pickle deserialization (#1104)
• refactor: RLM remove dead code, harden tunnels (#1107)
• refactor: RLM improve prompts and metrics (#1102)

Full Changelog: v0.1.12.dev2...v0.1.12.dev3