Skip to content

minimax m3 surges 198% – still priciest in a chinese-dominated top 4

Stone pillars showing AI model prices with towering M3 at $1.20 dwarfing cheaper V4 Flash, Hy3, and Mimo V2.5 models models

minimax m3 just grew 198% this week on openrouter – the biggest jump in the top 10

worth noting: they're running a -50% sale. and even at half price it's the priciest of the top 4 (all chinese models):

• deepseek v4 flash – $0.098 / $0.196
• tencent hy3 – $0.063 / $0.21
• xiaomi mimo v2.5 – $0.14 / $0.28
• minimax m3 – $0.30 / $1.20 (sale)

cheap pricing is eating the leaderboard

Ranked table of top 10 models by token usage on OpenRouter with MiniMax M3 at rank 3 showing 3.63T tokens and 198% weekly growth

what minimax m3 is

minimax released m3 around june 1. its weights went up on hugging face on june 2 under the minimax community license. the model card names four things that set it apart:

  • native multimodality – text, images and video are mixed from the first training step
  • 1m tokens of context – 5x the 200k of the previous model, m2.7
  • minimax sparse attention – 9x faster prefill and 15x faster decoding than m2 at 1m tokens, per the model card. the technical paper came out on june 11
  • size – about 428b parameters, with about 23b active per token

that week m3 processed 3.63t tokens on openrouter and took third place in the top 10.

the price behind the jump

the sale put m3 at $0.30 per million input tokens and $1.20 per million output tokens. the full price was $0.60 / $2.40.

even at half price, m3 cost 4 to 6 times more per output token than the other three models in the top 4. all three were chinese models priced under $0.30 per million output tokens.

update, october 16, 2026: what changed since june

four months later, m3 is no longer in openrouter's weekly top 20. the table compares the four models from this post in june and today.

modelprice in june, in / outprice on oct 16, in / outweekly rank, data to oct 15
deepseek v4 flash$0.098 / $0.196$0.091 / $0.182#13, 3.11t tokens
tencent hy3$0.063 / $0.21$0.132 / $0.528#12, 3.26t tokens
xiaomi mimo v2.5$0.14 / $0.28$0.14 / $0.28not in the top 20
minimax m3$0.30 / $1.20 (sale)$0.30 / $1.20not in the top 20

the sale never ended. minimax's pricing page now marks the $0.30 / $1.20 rate as a permanent 50% discount. above 512k input tokens the price doubles to $0.60 / $2.40.

the cheapest way to run m3 on openrouter today is coreweave, at $0.23 / $0.96. its output price is still 3 to 5 times that of deepseek v4 flash or mimo v2.5.

all five models in this week's top 5 are chinese. deepseek v4.1 flash leads with 37.1t tokens, followed by step 5 preview, glm 5.3 flash, mimo-v2.6-flash and hy4 preview. xiaomi's newer mimo-v2.6-flash replaced mimo v2.5 at the top.

price still matters, but it no longer explains the whole chart. three of this week's top 5 cost less than m3. step 5 preview and hy4 preview cost more.

sources

Portrait of Addy Crezee, author at thehype.news

Addy Crezee

Founder, CEO & Editor-in-Chief at thehype. Addy is a serial entrepreneur building products and communities since 2014. In 2024, he founded /function1, one of the largest AI conferences in the world, scaling from 0 to 10k attendees in its first year.

More articles by Addy Crezee →
ON AIR · RADIO.THEHYPE.NEWS ↗ ai news radio — 24/7