LOADING DATASHEET
LOADING DATASHEET
by MiniMax
RANKED #117 OF 337 ASSISTANTS · OVERALL #187 OF 528 RANKED · VIBE SCORE 6.0 · 640 VOICES
COPY BADGE pastes a README snippet with the rank SVG.
MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a 1M-token context window, and is suited for long-horizon agentic work, coding,...
78 mentions
34 mentions
26 weeks · 870 voices
QUIETLY DEGRADING
Recent vibe and chatter are both sliding.
Open weights. You can use a hosted service, or download it and run it yourself, free.
Open weights: download MiniMax M3 once and it's yours. No subscription, no rate limits, works offline.
428B total parameters (23B active MoE), a 4-bit quant is about 208 to 265 GB
Apple Mac Studio M2 Ultra with 192 GB Unified Memory
192 GB UNIFIED MEMORY
~$5,500 NEW / $4,200Experimental 1-bit/2-bit GGUF (UD-IQ1_M or UD-IQ2_XXS), ~5 to 10 tok/s, 8k context. Too big for single consumer GPUs
Dual AMD Instinct MI300X or 2x NVIDIA RTX 6000 Ada Workstati
192 TO 384 GB VRAM
~$16,000 USEDINT4 / MXFP4 quant, ~20 to 30 tok/s, 32k to 64k context
4x NVIDIA H100 SXM5 Server Node
320 GB VRAM
~$120,000 SERVER BUILDFP8 / BF16 precision, 40+ tok/s, up to 1M context
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
Everyday jobs on each of those rigs: how long MiniMax M3 takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMApple Mac Studio M2 Ultra with 192 GB Unified Memory | SWEET SPOTDual AMD Instinct MI300X or 2x NVIDIA RTX 6000 Ada Workstati | FULL POWER4x NVIDIA H100 SXM5 Server Node |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~67 S★★★★★ | ~20 S★★★★★ | —★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~3.3 MIN★★★★★ | ~60 S★★★★★ | —★★★★★ |
| Build a websitea small landing page, markup and styles together | ~18 MIN★★★★★ | ~5.3 MIN★★★★★ | —★★★★★ |
| Build a backendan API with routes, storage and tests | ~56 MIN★★★★★ | ~17 MIN★★★★★ | —★★★★★ |
| Build a gamea playable browser game in one file | ~33 MIN★★★★★ | ~10 MIN★★★★★ | —★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. A DASH MEANS NO SPEED WAS RESEARCHED FOR THAT RIG. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
The internet is split on MiniMax M3. Biggest praise: price. Biggest gripe: saying no too much.
544 POSTS · 96 COMMENTS · REDDIT · GITHUB · X · BLUESKY · DEV.TO · LEMMY · DEV FORUMS · BLOG
10 thumbs up · 2 thumbs down
7 thumbs up · 3 thumbs down
0 thumbs up · 9 thumbs down
“config(seats): Nish lane refresh — cheapest+best of Qwen3.8/GLM5.3/DSv4 flash on openrouter+zenmux; free MiniMax M3 via commandcode/opencode.”
Checked picks first: tags say why you would switch, stars say how fully each one stands in for MiniMax M3.
Uh.. Honey, how do you feel about takeout?
Please god, please there will be a smaller variant that also come with it
MiniMax M3 is out now!
Minimax M3 support with MSA has been merged into llama.cpp
About Xiaomi and their censoring hiccup
Deepseek V4 pro vs Minimax M3. Judge is Opus 4.8. Results are disappointing
Minimax M3 has been released
PlotPoints NSFW RP Voting Arena Now Live! (SFW version updated) 40 models up from 21; please go vote!
MiniMax M3 launched!
Gemini Flash 3.7 is at 50% discount on OpenRouter
Why doesn’t GitHub Copilot support open-weight models now that pricing is token-based?
How are you all speeding up Minimax M3 generations?