MiniMax M3.1 Flash + SGLang Omni: building toward a more interactive generation of voice agents. The shift ahead is from “ask, wait, listen” to conversations people can interrupt, correct, and redirect as a task unfolds.
MiniMax M3.1 Flash brings intelligence and fast, concise adds streaming, conversation state, and interruption handling so users can redirect conversations in real time.
Together, they give developers a foundation to explore proactive tutors, support agents, and hands-free copilots - pairing Flash’s capabilities with infrastructure for continuous interaction.
Thanks to the SGLang Omni team for exploring this direction with M3.1 Flash! ✨
MiniMax M3.1 Flash brings intelligence and fast, concise adds streaming, conversation state, and interruption handling so users can redirect conversations in real time.
Together, they give developers a foundation to explore proactive tutors, support agents, and hands-free copilots - pairing Flash’s capabilities with infrastructure for continuous interaction.
Thanks to the SGLang Omni team for exploring this direction with M3.1 Flash! ✨
We just tried @MiniMax_AI's M3.1 Flash in our SGLang Omni cascaded voice interaction system, and the experience was genuinely impressive! 🐰🚀
Our original pipeline was Qwen3-ASR → Qwen3 LLM → Qwen3-TTS. We replaced the locally hosted LLM with the MiniMax M3.1 Flash API while pipeline unchanged, and recorded a demo to see how it performs in real conversations.
The results were excellent. M3.1 Flash responded incredibly fast, understood our questions well, and consistently delivered sharp, concise, and direct answers. The conversation felt remarkably smooth and natural.
One thing I increasingly appreciate about voice agents is that intelligence alone isn't enough. Unlike text, users can't easily skim an audio response. A model that understands what you need and gets straight to the point can make a huge difference to the overall interaction experience.
This is also what makes the modular architecture of SGLang Omni so exciting to me. We can swap individual components, connect local speech models with frontier LLM APIs, and immediately explore how different models change the end-to-end interaction experience.
Huge congrats to the MiniMax team! Really impressive work, and we're excited to explore more together. 🔥
22
18
560
MiniMax M3.1 Flash: Space Performance + Bunny Speed
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence s Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
🚀🐰 Yes, Space Bunny - which has topped global leaderboards with 200T+ tokens - was an earlier version of M3.1 Flash.
💰Redefining the performance economics of agentic AI. M3.1 Flash achieves frontier-level intelligence s Cowork (Automation-Bench: 62%; ALE: 30%), Coding (DeepSWE: 74%) and Visual Reasoning (BabyVision: 91%), at 1/3 the cost of M3 and 3x the speed.
API pricing (flat rate): $0.15 / $0.60 per 1M input/output tokens, with cache reads at $0.015 / 1M tokens.
Model weights will arrive on Oct 23.
API: platform.minimax.io
M Plan: minimax.io/m-plan
MiniMax Code for desktop: agent.minimax.io/download
466
674
7,483
New York, you brought the energy! 🗽💜
At #AWNewYork26, MiniMax brought together @krea_ai , @fal & @magnific for a panel full of sharp insights and fresh perspectives.
As we deepen our work with brands and agencies, we explored what makes AI video ready for production: conse control, and work that feels true to the brand.
Our booth was buzzing, and our little octopus & tote bags got plenty of love 🐙
Thanks for having us, @advertisingweek!
At #AWNewYork26, MiniMax brought together @krea_ai , @fal & @magnific for a panel full of sharp insights and fresh perspectives.
As we deepen our work with brands and agencies, we explored what makes AI video ready for production: conse control, and work that feels true to the brand.
Our booth was buzzing, and our little octopus & tote bags got plenty of love 🐙
Thanks for having us, @advertisingweek!
10
3
305
MiniMax's President of Global Business, Linda Sheng, spoke today at the inaugural DBS Asia Future Forward Conference 2026 in Singapore. 🇸🇬
Alongside her was a distinguished lineup of government and business leaders, including Singapore Prime Minister Lawrence Wong and TikTok CEO ntribute MiniMax's perspective to the conversation on Asia's future. Thank you to @dbsbank for having us!
#DBSFutureForward
Alongside her was a distinguished lineup of government and business leaders, including Singapore Prime Minister Lawrence Wong and TikTok CEO ntribute MiniMax's perspective to the conversation on Asia's future. Thank you to @dbsbank for having us!
#DBSFutureForward
20
8
577
🪩 Customer Spotlight: SILVERSIDE · “We Bring the Fire”
Our longtime creative partner & client brings bold imagination, beautiful craft, and irresistible charm to MiniMax Music + H3 animation and lip sync.
@Silverside_AI @maxado_ every frame has personality. Every beat brings the fire 🔥🎶
Our longtime creative partner & client brings bold imagination, beautiful craft, and irresistible charm to MiniMax Music + H3 animation and lip sync.
@Silverside_AI @maxado_ every frame has personality. Every beat brings the fire 🔥🎶
7
3
272
@friendliai @anacondainc @kilocode @novita_labs @ArtificialAnlys @NousResearch @radixark @vercel October 8th MiniMax Research Panel @FireworksAI_HQ @digitalocean @Modular @radixark
partiful.com/e/DKY1zYjyuOvM…
partiful.com/e/DKY1zYjyuOvM…
1
0
4
MiniMax at SF Tech Week 🌉
📍 Oct 6 · Frontier Models, High-Speed Inference & Agents at Scale With @friendliai , @anacondainc , and @kilocode.
📍 Oct 7 · Open Source AI Stack With @novita_labs, @ArtificialAnlys, @NousResearch , @radixark , and @vercel.
📍 Oct 8 · InteniMax research night with @FireworksAI_HQ , @radixark , @digitalocean , @Modular, and @evermind .
📍 Oct 8 · Magnific Tech: Growth at the Pace of AI MiniMax joins @magnific to discuss building and growing creative AI products.
See you in SF.
Details + RSVP 👇
📍 Oct 6 · Frontier Models, High-Speed Inference & Agents at Scale With @friendliai , @anacondainc , and @kilocode.
📍 Oct 7 · Open Source AI Stack With @novita_labs, @ArtificialAnlys, @NousResearch , @radixark , and @vercel.
📍 Oct 8 · InteniMax research night with @FireworksAI_HQ , @radixark , @digitalocean , @Modular, and @evermind .
📍 Oct 8 · Magnific Tech: Growth at the Pace of AI MiniMax joins @magnific to discuss building and growing creative AI products.
See you in SF.
Details + RSVP 👇
14
5
71
@advertisingweek @krea_ai @fal @magnific 🎙️ The New Creative Stack: AI Video, Brand Control & the Future of Advertising
w/ Helena Grau i Miarons (@krea_ai), Nina Gerov (@fal), Claire Xue (@magnific), @RenLeanna & @MorganSuoinMM (@MiniMax_AI)
📅 Thu 10/8 · 2:30 PM ET · Innovation Stage
🔗newyork2026.advertisingweek.com/aw/schedule/se…
w/ Helena Grau i Miarons (@krea_ai), Nina Gerov (@fal), Claire Xue (@magnific), @RenLeanna & @MorganSuoinMM (@MiniMax_AI)
📅 Thu 10/8 · 2:30 PM ET · Innovation Stage
🔗newyork2026.advertisingweek.com/aw/schedule/se…
1
1
167
New York, see you next week 🗽✨
MiniMax is heading to @advertisingweek New York, Oct 5–8.
📍 Booth A16E, near the Food Trucks
🎬 Live MiniMax H3 demos all week
🎙️ Our panel with @krea_ai × @fal × @magnific : Thu 10/8 · 2:30 PM ET · Innovation Stage
Panel details in the AWNewYork26
MiniMax is heading to @advertisingweek New York, Oct 5–8.
📍 Booth A16E, near the Food Trucks
🎬 Live MiniMax H3 demos all week
🎙️ Our panel with @krea_ai × @fal × @magnific : Thu 10/8 · 2:30 PM ET · Innovation Stage
Panel details in the AWNewYork26
8
4
340
We’re heading to @aiDotEngineer NYC! 🗽
Morgan Suo, MiniMax’s Head of US Business Development, will present on “The Hidden Costs of a Faster Model.”
They'll cover how quantization, speculative decoding, and reasoning controls change a model’s cost, speed, and quality, and itching.
📅 October 14, 2026 · 10:40–10:58 AM ET
📍 McCarthy Room · Sheraton New York Times Square Hotel
See you there!
Morgan Suo, MiniMax’s Head of US Business Development, will present on “The Hidden Costs of a Faster Model.”
They'll cover how quantization, speculative decoding, and reasoning controls change a model’s cost, speed, and quality, and itching.
📅 October 14, 2026 · 10:40–10:58 AM ET
📍 McCarthy Room · Sheraton New York Times Square Hotel
See you there!
8
3
63
We're proud to contribute as an AI partner to the newly launched @Singtel AI Pass, supporting Singapore's SkillsFuture AI Subscription initiative. 🇸🇬
MiniMax H3 (@Hailuo_AI), MiniMax Agent (@MiniMaxAgent) and MiniMax Audio are all included, giving eligible learners across 200+ es hands-on access to premium AI tools and the opportunity to build practical, real-world AI skills.
This partnership brings our mission, "Intelligence with Everyone", to life, empowering learners across Singapore to explore, create, and build confidently with AI.
Learn more: singtel.com/personal/produ…
MiniMax H3 (@Hailuo_AI), MiniMax Agent (@MiniMaxAgent) and MiniMax Audio are all included, giving eligible learners across 200+ es hands-on access to premium AI tools and the opportunity to build practical, real-world AI skills.
This partnership brings our mission, "Intelligence with Everyone", to life, empowering learners across Singapore to explore, create, and build confidently with AI.
Learn more: singtel.com/personal/produ…
11
6
168
Three paths to faster video attention: compute the same interactions more efficiently, compute fewer in full, or change how information is mixed. Here’s a visual guide. 👇
Thanks to Nunchux AI and collaborators for VC-Attention, bringing training-free low-bit acceleration to fidelity than SageAttention2 in the B200 evaluation.
The approach balances speed and fidelity: V-Smooth reduces value quantization error, while ExpCast-FP8 makes softmax faster through approximation.
Excited to see the community keep building on H3. Could combining low-bit computation with sparse methods like Sol-Attn push efficiency further? We’re looking forward to seeing that explored.
Thanks to Nunchux AI and collaborators for VC-Attention, bringing training-free low-bit acceleration to fidelity than SageAttention2 in the B200 evaluation.
The approach balances speed and fidelity: V-Smooth reduces value quantization error, while ExpCast-FP8 makes softmax faster through approximation.
Excited to see the community keep building on H3. Could combining low-bit computation with sparse methods like Sol-Attn push efficiency further? We’re looking forward to seeing that explored.
Introducing VC-Attention: fast and accurate low-bit attention without retraining.
On MiniMax-H3, VC-Attention speeds up attention by 1.6× on B200 and 1.5× on B300 over FlashAttention-4, with better fidelity than SageAttention2. It also works with existing sparse attention ations:
• V-Smooth reduces value quantization error.
• ExpCast-FP8 speeds up softmax.
Nunchux Attention, our proprietary extension, pushes the speedup to 1.9× on B200 and 1.8× on B300.
Blog: nunchux.ai/blog/attention…
Technical Report: arxiv.org/pdf/2609.15810
Joint work by researchers at MIT, CMU, UC Berkeley, Stanford, and NVIDIA.
20
35
329
The Japan IP AI Co-Creation Conference, co-hosted by KAGAMI AI and MiniMax, has officially concluded. 🇯🇵 🎉
We brought together more than 150 companies from Japan and the U.S., alongside distinguished guests including AKB48 producer Yasushi Akimoto, KAGAMI AI Chairman Takami tor and producer Motoi Chujo.
Global AI leaders @runwayml, @higgsfield, @krea_ai, and @HeyGen joined the conversation to explore the future of IP and generative AI.
We unveiled MiniMax H3 IP Edition, bringing the power of MiniMax H3 together with officially licensed Japanese IP for a new generation of AI-powered storytelling.
The conference received extensive coverage from major Japanese media, including a dedicated segment on TV Tokyo's WBS (@wbs_tvtokyo).
Japanese IP × Global AI.
A new era begins.
We brought together more than 150 companies from Japan and the U.S., alongside distinguished guests including AKB48 producer Yasushi Akimoto, KAGAMI AI Chairman Takami tor and producer Motoi Chujo.
Global AI leaders @runwayml, @higgsfield, @krea_ai, and @HeyGen joined the conversation to explore the future of IP and generative AI.
We unveiled MiniMax H3 IP Edition, bringing the power of MiniMax H3 together with officially licensed Japanese IP for a new generation of AI-powered storytelling.
The conference received extensive coverage from major Japanese media, including a dedicated segment on TV Tokyo's WBS (@wbs_tvtokyo).
Japanese IP × Global AI.
A new era begins.
25
12
995
Open weights. Shared progress. MiniMax H3 is moving fast.
We built H3 for video generation with native stereo audio and multimodal reference control. The open-source community is making that capability faster, more accessible, and easier to build on.
Recent highlights:
• a Lab and NVIDIA: 4-step distillation, now running on DGX Spark and Apple Silicon.
• Sol-H3 — NVIDIA’s SANA team: 15 seconds of 768p video + audio in 6.6 seconds on 8×B300, in the team’s warm-inference benchmark.*
• VDN — Haocheng Xi and the OpenVDN team: rethinking attention for faster H3 inference, with weights, training and inference code released.
• PDD — NVIDIA’s distillation method, brought to H3 by Alibaba PAI as 8-step Acc-LoRAs, now supported in ComfyUI.
• LightX2V — 4- and 8-step Turbo LoRAs, with workflows for text, image and reference-conditioned video + audio.
Behind every release are people training, optimizing, quantizing, testing and sharing. Special thanks to:
@haoailab @nuvalab @NVIDIAAI @xieenze_jr @HaochengXiUCB @ArashVahdat @julberner @LightX2V @ComfyUI
And to the individual contributors pushing the work forward:
@haozhangml @cxlcl1 @lawrence_cjs @yitongli165665 @haopengl33 @songhan_mit @shanasaimoe
Thank you for building with H3 and helping make it faster, more accessible, and more useful for the community. Powerful models go further when we build together. Keep pushing H3. Excited to see what comes next. 🚀
Explore the ecosystem: github.com/MiniMax-AI/awe…
We built H3 for video generation with native stereo audio and multimodal reference control. The open-source community is making that capability faster, more accessible, and easier to build on.
Recent highlights:
• a Lab and NVIDIA: 4-step distillation, now running on DGX Spark and Apple Silicon.
• Sol-H3 — NVIDIA’s SANA team: 15 seconds of 768p video + audio in 6.6 seconds on 8×B300, in the team’s warm-inference benchmark.*
• VDN — Haocheng Xi and the OpenVDN team: rethinking attention for faster H3 inference, with weights, training and inference code released.
• PDD — NVIDIA’s distillation method, brought to H3 by Alibaba PAI as 8-step Acc-LoRAs, now supported in ComfyUI.
• LightX2V — 4- and 8-step Turbo LoRAs, with workflows for text, image and reference-conditioned video + audio.
Behind every release are people training, optimizing, quantizing, testing and sharing. Special thanks to:
@haoailab @nuvalab @NVIDIAAI @xieenze_jr @HaochengXiUCB @ArashVahdat @julberner @LightX2V @ComfyUI
And to the individual contributors pushing the work forward:
@haozhangml @cxlcl1 @lawrence_cjs @yitongli165665 @haopengl33 @songhan_mit @shanasaimoe
Thank you for building with H3 and helping make it faster, more accessible, and more useful for the community. Powerful models go further when we build together. Keep pushing H3. Excited to see what comes next. 🚀
Explore the ecosystem: github.com/MiniMax-AI/awe…
81
91
907
Tokyo, this Friday 🇯🇵
MiniMax is bringing Japan's leading IP and entertainment voices together with some of the world's top generative AI companies.
Speakers: renowned producer Yasushi Akimoto, leaders from KADOKAWA, KAGAMIAI, @higgsfield, @krea_ai, @runwayml, and @HeyGen
📍 re you most excited to hear from? 👇
MiniMax is bringing Japan's leading IP and entertainment voices together with some of the world's top generative AI companies.
Speakers: renowned producer Yasushi Akimoto, leaders from KADOKAWA, KAGAMIAI, @higgsfield, @krea_ai, @runwayml, and @HeyGen
📍 re you most excited to hear from? 👇
25
19
152
Open models are changing the cost equation for production AI.
On September 16, MiniMax and @togethercompute are bringing Open by Design: The Economics of AI in Production to London 🇬🇧
We’ll get into:
⚙️ Where teams are finding real cost savings
🔀 How model selection and ng stack
📈 What it takes to scale open models without sacrificing performance or reliability
Joining the conversation:
🎙️ Rio Shen — GM of EMEA, MiniMax
🎙️ Max Ryabinin — VP of Model Shaping, Together AI
🎙️ Sarung Tripathi — VP of Customer Experience, Together AI
📅 September 16 | 6:30–8:30 PM BST
🍕 Pizza, drinks, and time to meet other builders
Registration 👇
On September 16, MiniMax and @togethercompute are bringing Open by Design: The Economics of AI in Production to London 🇬🇧
We’ll get into:
⚙️ Where teams are finding real cost savings
🔀 How model selection and ng stack
📈 What it takes to scale open models without sacrificing performance or reliability
Joining the conversation:
🎙️ Rio Shen — GM of EMEA, MiniMax
🎙️ Max Ryabinin — VP of Model Shaping, Together AI
🎙️ Sarung Tripathi — VP of Customer Experience, Together AI
📅 September 16 | 6:30–8:30 PM BST
🍕 Pizza, drinks, and time to meet other builders
Registration 👇
20
5
90
MiniMax H3 is at @anyscalecompute’s #RaySummit today in SF.
@VictorSuOrtiz will break down our 33B open-weight audio-video model at the @vllm_project Conference, bringing text, image, video + audio into one context.
@VoidAsuka will share how builders are fine-tuning, evaluating @digitalocean, @nvidia, @inferact, and @NousResearch.
From the model to the community building on it. See you there! 👋
🔗 Session details below:
@VictorSuOrtiz will break down our 33B open-weight audio-video model at the @vllm_project Conference, bringing text, image, video + audio into one context.
@VoidAsuka will share how builders are fine-tuning, evaluating @digitalocean, @nvidia, @inferact, and @NousResearch.
From the model to the community building on it. See you there! 👋
🔗 Session details below:
9
4
96
Love seeing real-world agent benchmarks like this 👇
MiniMax-M3: $0.018 to spin up an inbox, write and send a business email end-to-end. The cheapest model on the board - and it finished the job.
This is what efficient intelligence looks like.
(Yes! We’re a frontier lab - we and music models, we also train and open-source our frontier LLMs for you to use.)
MiniMax-M3: $0.018 to spin up an inbox, write and send a business email end-to-end. The cheapest model on the board - and it finished the job.
This is what efficient intelligence looks like.
(Yes! We’re a frontier lab - we and music models, we also train and open-source our frontier LLMs for you to use.)
56
24
469
The MiniMax H3 ecosystem is growing faster than ever! ☺️
Check out Awesome MiniMax H3 Integrations - an index tracking everything built around H3, from 24GB VRAM local ComfyUI setups to enterprise-scale SGLang & vLLM-Omni serving deployments.
🔥 What’s inside:
- Hardware & Precisioides across INT8, NVFP4, and GGUF down to 8GB VRAM limits
- Speed & Serving: Acceleration LoRAs, Sol-Attn, block-caching, and multi-GPU production stacks
- Developer Tooling: Native agent skills, timeline directors, multi-shot motion context nodes, and h3.c for Apple Silicon
Huge thanks to the global open-source community, framework maintainers, and independent developers pushing the boundaries of what’s possible with MiniMax H3!
👇 Explore the repository & contribute:
github.com/MiniMax-AI/awe…
Check out Awesome MiniMax H3 Integrations - an index tracking everything built around H3, from 24GB VRAM local ComfyUI setups to enterprise-scale SGLang & vLLM-Omni serving deployments.
🔥 What’s inside:
- Hardware & Precisioides across INT8, NVFP4, and GGUF down to 8GB VRAM limits
- Speed & Serving: Acceleration LoRAs, Sol-Attn, block-caching, and multi-GPU production stacks
- Developer Tooling: Native agent skills, timeline directors, multi-shot motion context nodes, and h3.c for Apple Silicon
Huge thanks to the global open-source community, framework maintainers, and independent developers pushing the boundaries of what’s possible with MiniMax H3!
👇 Explore the repository & contribute:
github.com/MiniMax-AI/awe…
37
41
482
We went a little over capacity at @magnific SF for MiniMax H3 😅🎬
From commercial work to music videos, we showcased how creators are already using H3 on Magnific.
Ethan Wei, AI Solutions Architect at @MiniMax_AI, gave an H3 walkthrough 🛠️ Claire Xue joined us for a fireside ative workflows are heading, and Enrique Lopez took H3 for a live spin inside Magnific. 🎥✨
Huge thanks to the Magnific team and everyone who packed the room ❤️
And if you want to play with H3 yourself: 2K generation is 50% off on Magnific through Sept 1. 👀
From commercial work to music videos, we showcased how creators are already using H3 on Magnific.
Ethan Wei, AI Solutions Architect at @MiniMax_AI, gave an H3 walkthrough 🛠️ Claire Xue joined us for a fireside ative workflows are heading, and Enrique Lopez took H3 for a live spin inside Magnific. 🎥✨
Huge thanks to the Magnific team and everyone who packed the room ❤️
And if you want to play with H3 yourself: 2K generation is 50% off on Magnific through Sept 1. 👀
19
3
121
🎵MiniMax-Music3
Next-Generation Open-Weights Production-Ready & Versatile Music Model
huggingface.co/MiniMaxAI/Mini…
Next-Generation Open-Weights Production-Ready & Versatile Music Model
huggingface.co/MiniMaxAI/Mini…
232
646
5,387
Tomorrow at 5:30pm PT: MiniMax H3 × @magnific at the Magnific SF office.
We’ll be showcasing what people are already making with H3, talking about where AI video is heading, and getting hands-on with H3 inside Magnific.
Expect a live workflow across cinematic generation, native g + more.
Then we’ll open things up for networking and more showcases from the community.
Only a couple spots left. Link below.
#MiniMaxH3 #AIVideo
We’ll be showcasing what people are already making with H3, talking about where AI video is heading, and getting hands-on with H3 inside Magnific.
Expect a live workflow across cinematic generation, native g + more.
Then we’ll open things up for networking and more showcases from the community.
Only a couple spots left. Link below.
#MiniMaxH3 #AIVideo
5
7
107
A few of the biggest takeaways from our H3 conversation on ThursdAI 🎥
Also, huge thank you to @altryne for having us and for bringing together such a great discussion.
🔓 Open video is reaching the frontier.
The discussion didn’t frame H3 as simply “good for an open model.” y alongside Seedance, FLUX and WAN, with @altryne calling H3 the “leading open weights open model.”
And the distinction was immediate: “Open weights, hostable by yourself, finetunable.”
⚡ Open weights compound incredibly fast.
Within ~48 hours of launch, the community had already brought LoRA support, Apple Silicon/MLX, ComfyUI quantization and optimizations across new hardware.
That’s exactly why we open-weighted H3: putting frontier capabilities in developers’ hands means the ecosystem can take the model places we never could alone.
🎭 Omni Reference + character consistency stood out as a major leap.
@blizaine called H3 “so flexible compared to a lot of the previous open-weight models” and said it was “as good as anything I’ve seen” for recreating a character across environments.
Images, voices, audio and video can all become references — making control and consistency increasingly as important as raw generation quality.
🛠️ Local and hosted workflows can complement each other.
We also discussed Context-IR + Regenerate-2K: generate locally with H3, optimize multimodal context when needed, then regenerate a 768p result at 2K using the original references rather than simply upscaling it.
📈 The broader video capability curve is moving incredibly fast.
@arena perspective summed it up well: “the changes we’ve seen in the fidelity and the quality and the sound is just unbelievable.”
Open models are no longer sitting on a separate curve — they’re increasingly competing at the frontier itself.
And a fitting note to end on from our host @altryne:
“Thank you guys for open weighting the models. We expect more.” 🙌
We hear you. 🔓
Also, huge thank you to @altryne for having us and for bringing together such a great discussion.
🔓 Open video is reaching the frontier.
The discussion didn’t frame H3 as simply “good for an open model.” y alongside Seedance, FLUX and WAN, with @altryne calling H3 the “leading open weights open model.”
And the distinction was immediate: “Open weights, hostable by yourself, finetunable.”
⚡ Open weights compound incredibly fast.
Within ~48 hours of launch, the community had already brought LoRA support, Apple Silicon/MLX, ComfyUI quantization and optimizations across new hardware.
That’s exactly why we open-weighted H3: putting frontier capabilities in developers’ hands means the ecosystem can take the model places we never could alone.
🎭 Omni Reference + character consistency stood out as a major leap.
@blizaine called H3 “so flexible compared to a lot of the previous open-weight models” and said it was “as good as anything I’ve seen” for recreating a character across environments.
Images, voices, audio and video can all become references — making control and consistency increasingly as important as raw generation quality.
🛠️ Local and hosted workflows can complement each other.
We also discussed Context-IR + Regenerate-2K: generate locally with H3, optimize multimodal context when needed, then regenerate a 768p result at 2K using the original references rather than simply upscaling it.
📈 The broader video capability curve is moving incredibly fast.
@arena perspective summed it up well: “the changes we’ve seen in the fidelity and the quality and the sound is just unbelievable.”
Open models are no longer sitting on a separate curve — they’re increasingly competing at the frontier itself.
And a fitting note to end on from our host @altryne:
“Thank you guys for open weighting the models. We expect more.” 🙌
We hear you. 🔓
The week open video caught the frontier. @blizaine is on ThursdAI right NOW breaking down all three, with a surprise: MiniMax's own @VictorSuOrtiz on stage for the H3 story.
MiniMax H3: open weights, #1 open video model on LMArena
FLUX 3 Video: BFL's first video model, native audio in a single pass
Seedance 2.5: native 30 second takes, 3 minute long videos
youtube.com/watch?v=FbKLj8…
10
6
94
Huge thanks to @ComfyUI for an incredible livestream!
From reference-to-video and native audio, to multi-shot commercial creation, MiniMax H3 showcased the expanding possibilities of open-weight video models. The stream also took a deep dive into quantization, offloading, al deployment on consumer-grade GPUs.
H3 is just getting started. We look forward to pushing creative boundaries alongside developers worldwide! Watch the full livestream ↓
From reference-to-video and native audio, to multi-shot commercial creation, MiniMax H3 showcased the expanding possibilities of open-weight video models. The stream also took a deep dive into quantization, offloading, al deployment on consumer-grade GPUs.
H3 is just getting started. We look forward to pushing creative boundaries alongside developers worldwide! Watch the full livestream ↓
Livestream Update: @MiniMax_AI H3: A New Open-Weight Video Model, Live in ComfyUI
MiniMax H3 is an open-weight, general-purpose multimodal video generation model that works across text, images, video, and audio.
In ComfyUI, you can use H3 for text-to-video, image-to-video, eneration, and reference-driven creation. H3 jointly generates the visuals and synchronized stereo audio, including dialogue, sound effects, ambience, and music, rather than adding audio afterward.
The open-weight H3 checkpoints support clips up to 15 seconds at 768p. MiniMax’s hosted H3 model also supports generation at up to 2K resolution.
During the stream, we’ll test the model live and discuss how H3 brings multiple generation tasks into one architecture, how its high-compression video representation improves efficiency, and what developers should know when setting it up locally through ComfyUI.
What we'll cover:
→MiniMax H3: open weights, native stereo audio, up to 2K/15s clips
→Live demo
→How H3 got small enough to run on consumer hardware
→ComfyUI workflow templates for H3
→Local workflow: use the H3 Context-IR API (platform.minimax.io/docs/api-..., generate locally at 768p, and use the H3 Regenerate-2K API
8
7
166
Thank you to everyone who joined our Reddit AMA! The community energy was incredible, and we loved diving deep into the architecture, workflows, and the future of MiniMax H3.🩵
For those who missed it, here is a full recap of what is shipping next and YES WE WILL KEEP OPEN UNTIL n-Source Mission
As copyright matters settle, transitioning to an Apache-2.0 license is on the table. We also owe you the technical details - a comprehensive technical report on H3's development is in the works and will be published soon~
🎥 Video & H3-Regenerate-2K
We are planning to open-source H3-Regenerate-2K. To be clear, this is a dedicated latent-space DiT regeneration model - not just a base checkpoint rerun and not a pixel upscaler. We are currently tuning its efficiency and quality so you can run it locally.
⚡ Architecture, Speed & Sparse Attention
Sparse Attention: Our sparse attention is MoBA-style, train-aware block selection. Expect a relatively conservative reference implementation in the near term with the goal of zero perceptible quality loss. Let’s build device-specific speedups together!
Low-Step Variants: The released checkpoint is already CFG-distilled. While we don't have a near-term commitment just yet, a 4-NFE / 8-NFE variant is under active consideration. In the meantime, huge shoutout to the community - the Turbo LoRAs you've built are genuinely fantastic.
🖼️ Unified Image Generation & Editing Single-frame
image generation was our one of most-asked topics (193 upvotes!). The answer is YES: we plan to open-source a unified text-to-image and general image-editing model derived directly from the H3 lineage. It is currently in post-training refinement.
⏱️ Pushing Limits: 60-Second Workflows
For longer videos, Ref2VA supports continuation (just feed the previous clip as the reference). And to the Redditor who chained together that 60-second workflow - great find! That capability is real and was retained from our pretraining phase~🫡
We can't wait to see what you build next. Let's keep pushing the boundaries of open AI together. 🚀
For details👇
For those who missed it, here is a full recap of what is shipping next and YES WE WILL KEEP OPEN UNTIL n-Source Mission
As copyright matters settle, transitioning to an Apache-2.0 license is on the table. We also owe you the technical details - a comprehensive technical report on H3's development is in the works and will be published soon~
🎥 Video & H3-Regenerate-2K
We are planning to open-source H3-Regenerate-2K. To be clear, this is a dedicated latent-space DiT regeneration model - not just a base checkpoint rerun and not a pixel upscaler. We are currently tuning its efficiency and quality so you can run it locally.
⚡ Architecture, Speed & Sparse Attention
Sparse Attention: Our sparse attention is MoBA-style, train-aware block selection. Expect a relatively conservative reference implementation in the near term with the goal of zero perceptible quality loss. Let’s build device-specific speedups together!
Low-Step Variants: The released checkpoint is already CFG-distilled. While we don't have a near-term commitment just yet, a 4-NFE / 8-NFE variant is under active consideration. In the meantime, huge shoutout to the community - the Turbo LoRAs you've built are genuinely fantastic.
🖼️ Unified Image Generation & Editing Single-frame
image generation was our one of most-asked topics (193 upvotes!). The answer is YES: we plan to open-source a unified text-to-image and general image-editing model derived directly from the H3 lineage. It is currently in post-training refinement.
⏱️ Pushing Limits: 60-Second Workflows
For longer videos, Ref2VA supports continuation (just feed the previous clip as the reference). And to the Redditor who chained together that 60-second workflow - great find! That capability is real and was retained from our pretraining phase~🫡
We can't wait to see what you build next. Let's keep pushing the boundaries of open AI together. 🚀
For details👇
📢 Official AMA Announcement
The complete MiniMax‑H3 development team will hold an Ask‑Me‑Anything session inside r/StableDiffusion.
The main researcher team will be here!
- dacongya (Head of H3 Researcher)
- Luigi (H3 Researcher)
- Nero (H3 Researcher)
- Kiro (H3 3 system engineer)
- Ryanlee (Head of Devrel)
Time:
PT 7:00 AM - 8:00 AM
Beijing: 22:00 - 23:00
Eastern Time: 10:00 - 11:00 AM
Central Europe: 16:00 - 17:00
reddit.com/r/StableDiffus…
57
66
751
Open weights = access → possibility.
Open communities. Better performance. Deeper customization. Faster innovation.
Thank you, @ComfyUI, @magnific, and @OpenRouter for backing H3 and building the open future with us. 💜
Open communities. Better performance. Deeper customization. Faster innovation.
Thank you, @ComfyUI, @magnific, and @OpenRouter for backing H3 and building the open future with us. 💜
The road to AGI was never meant to be traveled alone.
From Hailuo 01 to H3, we've been incredibly lucky to build alongside so many friends, partners, and fellow builders who believed in the same thing we did.
That's why we open-sourced H3. Because progress moves faster when more people can build on it, and because we want to keep moving forward together.
"To the edge of the sky" has always been more than a destination for us. It's a belief, and maybe a kind of idealism.
To everyone who has been part of this journey: thank you.
Together, toward the edge of the sky, we will keep going until we get there.🌌✨
11
13
1,076
Across infrastructure, creative platforms, and open communities, our partners share a belief in what open models can unlock.
Grateful to @fal, @pika_labs, and @HelloCivitai for building alongside us. 💜
#MiniMaxH3
Grateful to @fal, @pika_labs, and @HelloCivitai for building alongside us. 💜
#MiniMaxH3
The road to AGI was never meant to be traveled alone.
From Hailuo 01 to H3, we've been incredibly lucky to build alongside so many friends, partners, and fellow builders who believed in the same thing we did.
That's why we open-sourced H3. Because progress moves faster when more people can build on it, and because we want to keep moving forward together.
"To the edge of the sky" has always been more than a destination for us. It's a belief, and maybe a kind of idealism.
To everyone who has been part of this journey: thank you.
Together, toward the edge of the sky, we will keep going until we get there.🌌✨
5
6
956
Maestro v1.5.5 with MiniMax H3 is out! 🩵
The community got H3 running on hardware we never tested - from a gaming GPU to MacBooks running fully offline - and built tooling we didn't plan for within 48 hours.
We're genuinely glad we opened it up. <3
The community got H3 running on hardware we never tested - from a gaming GPU to MacBooks running fully offline - and built tooling we didn't plan for within 48 hours.
We're genuinely glad we opened it up. <3
Running MiniMax H3 locally for FREE is blowing my mind!
It's like having an unrestricted Seedance2.0/Sora2-level model you can run locally for FREE.
43
82
1,125
Claims that MiniMax H3 “cannot legally be used” in certain regions are incorrect.😅
MiniMax H3 can be licensed for deployment in the US, EU, UK, and South Korea through MiniMax’s formal authorization process. To apply, please submit the MiniMax H3 License Request Form found in and our team will process your request as soon as possible.👇
huggingface.co/MiniMaxAI/Mini…
Let's cook.🔥
MiniMax H3 can be licensed for deployment in the US, EU, UK, and South Korea through MiniMax’s formal authorization process. To apply, please submit the MiniMax H3 License Request Form found in and our team will process your request as soon as possible.👇
huggingface.co/MiniMaxAI/Mini…
Let's cook.🔥
We put explanations in the license that there are no regional restrictions. However, in the EU we follow conventions similar to LTX and Hunyuan; while in the US, due to local laws and regulations and the ongoing legal situation with Disney, users are required to complete an additional form and sign a waiver.
27
56
751
MiniMax-H3 Is Now Publicly Available
huggingface.co/MiniMaxAI/Mini…
huggingface.co/MiniMaxAI/Mini…
227
600
4,695
Introducing Intelligence in the Open 🧠🔬
A recurring MiniMax research series bringing together researchers and builders to share and discuss groundbreaking work across AI.
For our first edition on July 28, join us from 6:00–9:00 PM at UpScaleX in San Francisco, with talks in reasoning, agents, and evaluation from:
• Matt Turk, Senior ML Researcher at @withprotegeai — counterfactual evaluation for clinical LLMs and agents
• Federico Bianchi, Staff ML Scientist at @togethercompute — learning to discover at test time
• Will Keating, MTS at @ArtificialAnlys — AA-Briefcase, a frontier agentic knowledge-work evaluation
• Isha Chaturvedi, Founding Member at @KeenableAI — structured multi-agent reasoning and process-level error localization
Followed by an open panel discussion, food, and plenty of time to connect 🤝
Presented by MiniMax, @withprotegeai, @ArtificialAnlys and @togethercompute. Hosted at UpScaleX.
Hope to see you there!
Registration link in the reply 👇
A recurring MiniMax research series bringing together researchers and builders to share and discuss groundbreaking work across AI.
For our first edition on July 28, join us from 6:00–9:00 PM at UpScaleX in San Francisco, with talks in reasoning, agents, and evaluation from:
• Matt Turk, Senior ML Researcher at @withprotegeai — counterfactual evaluation for clinical LLMs and agents
• Federico Bianchi, Staff ML Scientist at @togethercompute — learning to discover at test time
• Will Keating, MTS at @ArtificialAnlys — AA-Briefcase, a frontier agentic knowledge-work evaluation
• Isha Chaturvedi, Founding Member at @KeenableAI — structured multi-agent reasoning and process-level error localization
Followed by an open panel discussion, food, and plenty of time to connect 🤝
Presented by MiniMax, @withprotegeai, @ArtificialAnlys and @togethercompute. Hosted at UpScaleX.
Hope to see you there!
Registration link in the reply 👇
12
10
95
MiniMax President and Co-founder Yeyi Yun met with H.E. Abdullah Alswaha and @McitGovSa to discuss deeper cooperation across generative AI, including text, audio, and video.
We look forward to building on this conversation and contributing to the continued growth of Saudi Arabia’s AI ecosystem.
We look forward to building on this conversation and contributing to the continued growth of Saudi Arabia’s AI ecosystem.
5
6
285
MiniMax is taking the stage @siggraph 2026.
On July 22, @RenLeanna, VP at MiniMax, will explore how native multimodal models connect understanding across modalities, and what this enables for richer AI systems and more capable creative workflows.
📍 Los Angeles
🕓 4:00–4:30 PM the comments👇
See you then!
On July 22, @RenLeanna, VP at MiniMax, will explore how native multimodal models connect understanding across modalities, and what this enables for richer AI systems and more capable creative workflows.
📍 Los Angeles
🕓 4:00–4:30 PM the comments👇
See you then!
7
9
279
Paris made one thing clear: open models are no longer a side conversation, they’re becoming central to how frontier AI is built, and scaled. 🇫🇷
Across RAISE week:
— Linda Sheng, President of Global Business at MiniMax, joined Luca Schmid, Managing Director HSG, on the Ada ss open models, and native multimodality, and the new enterprise use cases they make possible.
— RAISE House brought @togethercompute, @cast_ai, and @magnific together for a panel on open frontier models, applications, and infrastructure.
— Our private MiniMax Salon created space for deeper conversations with founders, developers, investors, and partners from @runware, @youdotcom, @ExaAILabs, and @FeatherlessAI
— Our reception with @SambaNovaAI, we brought together builders and infrastructure partners—including @InfercomAI, Argyll, and SCX ai—above Paris for live demos and discussions.
— At the booth, we met builders exploring how open models can move into production.
Thank you to everyone who joined us. Until next time, Paris.
Intelligence with Everyone. 🌍
Across RAISE week:
— Linda Sheng, President of Global Business at MiniMax, joined Luca Schmid, Managing Director HSG, on the Ada ss open models, and native multimodality, and the new enterprise use cases they make possible.
— RAISE House brought @togethercompute, @cast_ai, and @magnific together for a panel on open frontier models, applications, and infrastructure.
— Our private MiniMax Salon created space for deeper conversations with founders, developers, investors, and partners from @runware, @youdotcom, @ExaAILabs, and @FeatherlessAI
— Our reception with @SambaNovaAI, we brought together builders and infrastructure partners—including @InfercomAI, Argyll, and SCX ai—above Paris for live demos and discussions.
— At the booth, we met builders exploring how open models can move into production.
Thank you to everyone who joined us. Until next time, Paris.
Intelligence with Everyone. 🌍
4
6
144
July 9 - RAISE House with MiniMax.
A private executive gathering with:
Rio Shen, GM EMEA at MiniMax
Kai Mak, CRO at @togethercompute
Laurent Gil, Co-Founder & President at @cast_ai
Claire Xue, Global Head of AI Ecosystem at @magnific
Expect a conversation on frontier multdeployment.
luma.com/h1emgp78
A private executive gathering with:
Rio Shen, GM EMEA at MiniMax
Kai Mak, CRO at @togethercompute
Laurent Gil, Co-Founder & President at @cast_ai
Claire Xue, Global Head of AI Ecosystem at @magnific
Expect a conversation on frontier multdeployment.
luma.com/h1emgp78
0
0
38
July 8 - Linda Sheng, President of Global Business at MiniMax, joins Luca Schmid, Managing Director at HSG, on the Ada Lovelace Stage for “MiniMax: Scaling Intelligence for Everyone.”
A discussion on MiniMax’s vision, M3, open weights, and how we’re helping shape the next phase native multimodality.
luma.com/s1y0q6gr
A discussion on MiniMax’s vision, M3, open weights, and how we’re helping shape the next phase native multimodality.
luma.com/s1y0q6gr
0
0
37
