TMax v1.1 Released, Tops Open-Source Terminal Agent Benchmark After Data Fixes

WAI released TMax v1.1, an RL training recipe for terminal agents. After auditing revealed that 87% of v1.0's training environment validators were flawed, fixes to data and infrastructure boosted Terminal Bench 2 scores by up to 6 points, with the 9B model surpassing Claude Haiku 4.5 and leading open-source terminal agents under 30B.

2026-11-03 ~ 2026-11-04 · 3 related posts

• TMax v1.1 released: open terminal agents under 30B gain up to 6 points on TB2 — natolambert · 2026-11-03
• TMax v1.1 fixes flawed RL data—87% of environments had weak verifiers—and beats Claude Haiku 4.5 with a 9B model — RulinShao · 2026-11-03

1 near-duplicate retellings: RulinShao