LMArena actually has a nice Pareto distribution of ELO vs price for this

  model                        elo   $/M
  ---------------------------------------
  glm-5.1                      1538  2.60
  glm-4.7                      1440  1.41
  minimax-m2.7                 1422  0.97
  minimax-m2.1-preview         1392  0.78
  minimax-m2.5                 1386  0.77
  deepseek-v3.2-thinking       1369  0.38
  mimo-v2-flash (non-thinking) 1337  0.24

qwen3.5/3.6 (30B) works well,locally, with opencode

GLM 5.1 via an infra provider. Running a competent coding capable model yourself isn't viable unless your standards are quite low.

I am using GLM 5.1 and MiniMax 2.7.