Back to News
SourceMarsBit

MiniMax携M3、H3等全模态模型及产品亮相WAIC2026

火星财经消息,7 月 17 日,在 2026 世界人工智能大会(WAIC2026)期间,MiniMax 集中展示旗舰模型 MiniMaxM3,以及多模态模型和 AI 原生产品,其中新一代多模态生成模型 H3 在 WAIC 期间抢先预热。通过机器狗、AI 眼镜、智能耳机、NAS、金融服务和影视内容等生态合作案例,MiniMax 展示了 AI 从底层模型、智能体产品到具体应用的发展与落地实践。 此次展示覆盖模型能力、学术研究、产品工具、多模态创作和行业应用等多个层面。观众可以了解 M3 在长上下文、Coding 和 Agentic 任务上的能力,也可以体验多模态模型在创作工作流、智能硬件和行业服务中的具体应用,并了解 MiniMax 在开源生态、产业合作和全球化方面的进展。(澎湃)
Disclaimer: The views above are the author's only and do not represent 711BTC. Nothing here constitutes investment advice.

Related

06-18 15:05

B.AI Introducing a 7-day limited-time free trial of the MiniMax-M3

ChainCatcher reports that MiniMax M3, which topped the Artificial Analysis open-source leaderboard, is offering a 7-day limited-time free trial. This open-source model, which integrates native multimodality, millions of context records, and top-tier coding inference capabilities, is now freely available to all users. Web users can log in to use the model directly for free, and developers can create an API key through the API management panel to call it. Exclusive benefits are also available: new users who log in using Binance Wallet or Bitget Wallet can receive 1 million free Credits and enjoy bonus points on top-ups (BNB Chain 1:1, others 1:0.5), with a maximum additional reward of $100 per user. As a one-stop platform that aggregates leading global AI, B.AI is the first to bring users a zero-threshold experience with MiniMax M3. Whether it is complex code library analysis, multimodal content generation or ultra-long document processing, it can be easily handled, truly making top open source capabilities accessible.

06-12 23:11

MiniMax M3 has been officially open-sourced, with native multimodal support for millions of contexts.

According to Beating, a Chinese large-scale model manufacturer, MiniMax has officially open-sourced the weights of its native multimodal hybrid expert (MoE) model, MiniMax M3, on Hugging Face. MiniMax M3 boasts a total of 428 billion parameters, with 23 billion parameters activated by a single token, and natively supports over 1 million contexts. To reduce deployment memory overhead, the development team simultaneously released an MXFP8 quantized version, adapting it to mainstream inference frameworks such as SGLang, vLLM, and Transformers. In its multimodal design, MiniMax M3 performs joint training of text, images, and videos during the pre-training phase to achieve native semantic fusion, rather than performing multimodal alignment in the post-training phase. In terms of its operational mechanism, the model provides dual inference modes: a Thinking mode for complex logic and tool orchestration, and a Non-thinking mode for low-latency dialogue and code generation. The underlying kernel supporting over 1 million contexts is the simultaneously open-source lightweight attention kernel library, MiniMax Sparse Attention (MSA). Official data shows that MSA uses a grouped query attention (GQA) block retrieval mechanism. In a real-world test with an extremely long context of 1 million tokens, the MSA operator optimized for the NVIDIA Blackwell (SM100) architecture can achieve more than 9 times the pre-filling speedup and 15 times the decoding speedup compared to the traditional full attention mechanism, while significantly reducing inference overhead.

07-08 17:10

One hour before Trump's statement, a short position of 21 million (long oil, short Nasdaq) saw significant profits after investors added to their short positions.

According to BlockBeats, on July 8th, Hyperinsight monitoring showed that an address starting with 0xec4 had already established a long position in Brent crude oil with 20x leverage two days prior. Then, approximately one hour before Trump's Middle East statements triggered a market surge, a short position in the Nasdaq 100 was significantly increased with 30x leverage. Subsequently, oil prices jumped, and risk assets came under pressure, leading to profit-taking on both sides. As of press time, the two highly leveraged positions at this address had a combined unrealized profit of approximately $510,000. The return on the Brent crude oil long position reached 189%, and the return on the Nasdaq 100 short position reached 37%. The entry times were as follows: July 6th afternoon: Brent crude oil long position established with 20x leverage; July 7th evening: Nasdaq 100 mapped contract (XYZ100) opened with a base position of approximately 200 contracts with 30x leverage; This afternoon (approximately 1 hour before the statement): Significantly increased short positions, pushing the XYZ100 short position to 600 contracts, valued at approximately $17.304 million, completing the setup just before the news was released. On the news front, Trump's latest statement indicated that the temporary ceasefire arrangements between the US and Iran may have ended, and the market re-priced in Middle East geopolitical risk premiums. Reports indicate that crude oil futures jumped approximately 5% after the statement. Current core positions: Brent crude oil (BRENTOIL) long positions: 50,000 contracts, approximately $3.915 million, average price $71.8645, liquidation price $37.26, unrealized profit approximately $322,000; Nasdaq 100 mapped contract (XYZ100) short positions: 600 contracts, approximately $17.304 million, average price $29,108.44, liquidation price $32,119.96, unrealized profit approximately $161,000.

07-08 16:57

ZTE's world's first AI-powered smartphone will be unveiled at WAIC 2026.

According to Mars Finance, Blue Whale Technology reporters learned from sources that ZTE Nubia's world's first AI-powered smartphone may officially debut at the 2026 World Artificial Intelligence Conference (WAIC). (Blue Whale News, Zhai Zhichao)

07-08 16:37

MiniMax plans to launch a large model with 2.7 trillion parameters.

PANews reported on July 8th that, according to sources, AI company MiniMax is developing a new large-scale language model with 2.7 trillion parameters, exceeding the scale of all currently available domestic AI models. The sources stated that this new model is expected to be released as early as the third quarter of this year. Internal employees involved in the development have designated it as M3 Pro, but it is unclear whether the company will use this name in the official release. MiniMax plans to open-source the model. This new model is significantly larger than MiniMax's current flagship model, M3 (428 billion parameters). Larger-scale AI models are better suited for handling complex reasoning and multi-step instruction tasks.

07-08 16:33Important

S&P 500 E-Mini futures fell 0.7%, and Nasdaq 100 futures fell 1.1%.

According to Mars Finance, Gate market data shows that S&P 500 E-Mini futures fell 0.7%, Nasdaq 100 futures fell 1.1%, and Dow Jones futures fell 0.9%.