1# Model family selection
 2
 3Prefer diverse underlying model families, not just diverse providers or
 4gateways. A model target such as `hyper/minimax-m3` has provider/gateway
 5`hyper`, family MiniMax, and model M3.
 6
 7## Default preference
 8
 9Prefer open-weight or otherwise cost-effective non-frontier-proprietary models
10unless the user pins a different target. Avoid known-proprietary labs like
11Anthropic or OpenAI by default because they are expensive and not necessary for
12the intended review mix.
13
14Generally prefer review lanes from these particular models, or newer/larger,
15when available:
16
17- Qwen 3.7 Plus, preferably Max
18- GLM 5.1
19- Kimi K2.6
20- MiniMax M3
21- Mimi V2.5 Pro
22- DeepSeek V4 Pro
23
24When multiple variants in a generation are available, generally prefer the
25largest or strongest available variant: Max over Plus over Flash, or Qwen 3.6
26397B over Qwen 3.6 35B.
27
28## Roster fields
29
30Record these for every active or skipped lane:
31
32- runner
33- runner invocation profile
34- exact candidate model ID
35- provider/gateway
36- underlying model family
37- generation
38- variant
39- intended role
40- reason selected or skipped