[Feature Request] Reconsider MoE architecture - performance regression compared to Opus 4 variants
Status Open
Reported on v2.1.224
Maintainer reply None cached
Activity 0 comments · opened Aug 11, 2026
Bug Description
stop do a MoE model it stupid it cant compare with opus-4-8 opus4-6 opus4-2 is ten time better rn you implement to be like GPT worst than OG-opus , it feel like cant trust anymore no mind , cantg understand big problem , people are going to leve you for real if you keep going smaller size Active Parameter , i dont know how much active parameter rn but it did feel liek 30-60 not more than this it cant work with this ,
you can see by result/token result/time use , it took so much time like 4X longer to get to the problem correct , i dont care for how durable agentic you claim
Environment Info
- Platform: darwin
- Terminal: ghostty
- Version: 2.1.224
- Feedback ID: 7aa852c8-fb61-4612-a70f-dcb52990947a
Errors
[]