minimax-m3.1POST
Space Bunny trained on 200T tokens in stealth; M3.1 Flash claims agentic and computer-use gains
The poster asks whether operating anonymously let MiniMax iterate faster without VC and enterprise feedback loops, or whether this is simply how non-OpenAI labs build models now.
gianluca
@glucapavArchived 7 days agoOriginally posted 2026.10.20
gianluca @glucapavOriginally posted 2026.10.20
minimax trained space bunny on 200T tokens in stealth mode, now releasing m3.1 flash claiming gains on agentic tasks and computer use. the interesting question: did operating anonymously let them iterate faster without the usual vc/enterprise feedback loops slowing things down, or is this just how you build models now when you're not openai
x.com/glucapav/status/2112242673184641104Archived 2026.10.20 20:42