MiniMax M3.1 Flash: $0.15 AI Model, Honestly Tested

MiniMax just revealed that "Space Bunny", the anonymous model that climbed OpenRouter's charts, was an early version of MiniMax M3.1 Flash. I put it through a real coding test, told as a late-night story between a tired developer and Space Bunny. What MiniMax claims: $0.15 / $0.60 per 1M input/output tokens; cached reads $0.015 (flat rate) 1/3 the cost of M3 and 3x the speed Benchmarks (self-reported): BabyVision 91%, DeepSWE 74%, Automation-Bench 62%, ALE 30% "200T+ tokens" served as Space Bunny (cumulative; OpenRouter's own count was 56.7T in 30 days, #2 by volume) Open weights on October 23; 1M-token context My test (OpenRouter, streamed, same task for each model): write a CSV parser (quoted fields, escaped quotes, newlines inside fields), then I ran each model's code against 12 test cases I wrote. M3.1 Flash: passed all 12 tests in 3 of 3 runs · ~104 tokens/s streaming · but it "thinks" 5,000–11,000 tokens first, so each answer took ~100 s NVIDIA Nemotron 3 Super (free): 1 of 3 runs passed · ~36 tokens/s Google Gemma 4 31B (free): rate-limited 3 of 3 times, no answer On a first round capped at 4,000 tokens, Flash ran out of room before writing code; give it a big token budget In that first round, M3 streamed at about the same token speed as Flash (91–123 tok/s), so I couldn't confirm "3x the speed" in raw streaming speed Free-tier runs over my own connection; speeds vary by provider and time. Chapters 0:00 2 AM 0:32 Who is Space Bunny 1:03 Their benchmarks 1:21 My test 2:26 Fine print + verdict Sources (checked 19 Oct 2026) MiniMax announcement: https://x.com/MiniMax_AI/status/21121... MarkTechPost: https://www.marktechpost.com/2026/10/... FourWeekMBA (OpenRouter 56.7T / #2): https://fourweekmba.com/ai-minimax-pr... OpenRouter model list (pricing, context, free launch variant) My own test over OpenRouter #MiniMax #AI #LLM #AIModels #OpenRouter #AICoding #SpaceBunny