Rolling out October 19, 2026, M3.1 Flash is MiniMax's next "Flash" tier — tuned for tool use, coding, and multi-step automation, with output above 200 tokens per second and pricing near the market floor. The model that ran anonymously as "Space Bunny" was an early version.
2026-10-19 · by Moe Ameen
On October 19, 2026, MiniMax began rolling out M3.1 Flash, the newest model in its "Flash" tier and one it positions squarely for agentic work — tool use, coding, computer use, and multi-step automation across enterprise systems. The Shanghai company says the model that had been circulating anonymously as "Space Bunny," which processed more than 200 trillion tokens, was an early version of M3.1 Flash. It is available now through the MiniMax API, the M Plan, and MiniMax Code, where it had been in preview since late September.
The headline pitch is speed and price. MiniMax prices it flat at $0.15 per million input tokens and $0.60 per million output, with cached input at $0.015 — and, per third-party reporting, no long-context surcharge, unlike the step-up pricing on MiniMax M3. It cites everyday output speeds above 200 tokens per second (a median around 170), roughly three times faster online inference than before, and about 70% lower cost on agentic tasks than its predecessor. Multiple trackers list a context window around one million tokens; the model reads text and images and returns text.
On MiniMax's own numbers the model posts 76.2 on BankerToolBench, 73.7% on DeepSWE, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82% on CyberGym, and MiniMax claims it ranks first among Chinese models on OSWorld V2 and ALE. The independent Artificial Analysis Intelligence Index puts it at 42 — 13 points above MiniMax M3 at less than half the cost per task, level with GLM-5.3-Flash and behind DeepSeek V4.1 Flash 1017 — while flagging a high hallucination rate on its knowledge benchmark, a reminder that the model is tuned for cheap agentic throughput rather than factual recall. MiniMax has not confirmed a parameter count (some reports cite figures carried over from M3), says open weights will follow later, and promises a more capable "Pro" model to come.
The useful way to read a launch like this is not "which cheap model should I switch to" — it is that raw text generation is no longer where a creator's time or money goes. When a million tokens of competent drafting costs cents and streams at 200 a second, the bottleneck moves downstream, to the work M3.1 Flash explicitly does not do: turning words into finished media and getting them in front of people. That downstream half is what [Kompozy](/) is built for.
Concretely, if you want to exploit cheap, fast reasoning, use a model like this to draft in bulk or distill a long source, then hand the result to Kompozy as a seed. Kompozy fans one idea into formats the model can't render — a captioned [Persona Shorts](/glossary/persona-shorts) avatar video, a brand-exact [Carousel](/glossary/hyperframes), Quote Graphics, per-platform Text Posts, a [Blog Article](/ai-tools/koala-ai), and an Email Newsletter — all held to one voice by a [Persona Brief](/glossary/persona-brief) with banned-word filters, which also screens the robotic phrasing and backstops the model's weakest trait by keeping a human review gate on every post. Then [Autopilot](/glossary/autopilot) schedules and publishes the set across the eight social platforms plus blog and email. (Kompozy's own copy runs on Claude and OpenAI, so this is about the shared workflow, not a model swap.) The cheap model makes drafting nearly free; Kompozy turns a free draft into a published week.
It is MiniMax's fast, low-cost model built for agentic tasks — tool use, coding, computer use, and multi-step automation — that began rolling out on October 19, 2026 through the MiniMax API, M Plan, and MiniMax Code. It reads text and images and returns text, and MiniMax says it runs about three times faster than its previous model.
MiniMax lists $0.15 per million input tokens and $0.60 per million output, with cached input at $0.015 per million, charged flat with no long-context surcharge according to third-party reporting. MiniMax also claims roughly 70% lower cost on agentic tasks than its predecessor. Rates change, so confirm on MiniMax's pricing page.
It is good at drafting text fast and cheap, but it is tuned for agentic work and independent trackers flag a high hallucination rate, so treat its output as a draft to fact-check. It generates no video, images, or posts and publishes nothing — for finished, multi-platform content you pair its drafts with an engine like Kompozy.