Announcement
Tev-Gym announced with 1,170 reinforcement-learning environments
Its creators say Tev-27B, trained on Tev-Gym, matches GPT-6 Astra and Claude Opus 5 on ALE-CM at Pass@5.
TLDR
Tev-Gym's creators say its 1,170 environments offer dense progress rewards for long-horizon agentic reinforcement learning. They report that Tev-27B matches GPT-6 Astra and Claude Opus 5 on ALE-CM, covering computing and math, at Pass@5. They also say Tev-320B tops open models on Terminal-Bench 4.0-CM at Pass@1, trailing GPT 5.6 sol by 0.02.
Combined views
2
3 Sources, first seen ago
