Terminal-Bench 2.0 on DeepSeek-Flash
89 tasks, 1 attempt, allowlisted network.
One attempt per task is noisy, so read close scores as a tie. Bring your own DeepSeek key.
89 tasks, 1 attempt, allowlisted network.
One attempt per task is noisy, so read close scores as a tie. Bring your own DeepSeek key.