Models Can Also "Nest"? MiniMax Releases M2.7: The First Domestic Large Model Deeply Involved in Self-Iteration

marsbit2026-03-18 tarihinde yayınlandı2026-03-18 tarihinde güncellendi

Özet

Artificial intelligence is evolving from monthly updates to self-evolution. On March 18, MiniMax released its first new model version deeply involved in its own iteration—MiniMax M2.7. This marks a new stage in model development: large models are no longer solely trained by human programmers but have begun to "train themselves." The core breakthrough of MiniMax M2.7 lies in its strong autonomous construction capability. It can independently build complex Agent Harness (intelligent agent testing frameworks) and, relying on underlying capabilities such as Agent Teams, complex Skills, and Tool Search tools, complete highly complex productivity tasks autonomously. In simple terms, M2.7 is not just a smarter conversational agent but also a "digital engineer" capable of self-diagnosis and self-optimization. This "self-participatory iteration" model will significantly enhance the model’s logical reasoning and tool invocation accuracy when facing unknown complex tasks. Currently, this self-evolving MiniMax M2.7 model has been fully launched on the MiniMax Agent platform and open platform. As large models begin to deeply participate in their own "growth" process, the ceiling of AI may be raised once again.

The evolution speed of artificial intelligence is transitioning from "monthly updates" to "self-evolution." On March 18, MiniMax officially released its first new version model deeply involved in iterating itself—MiniMax M2.7. This marks a new stage in model development: large models are no longer solely fed by human programmers but have begun to learn to "guide themselves."

According to reports, the core breakthrough of MiniMax M2.7 lies in its powerful autonomous construction capability. It can independently build complex Agent Harness (intelligent agent testing frameworks) and, relying on underlying capabilities such as Agent Teams (intelligent agent collaboration), complex Skills, and Tool Search tool, independently complete highly complex productivity tasks.

Simply put, M2.7 is not just a smarter conversationalist but also a "digital engineer" capable of self-diagnosis and self-optimization. This "self-participatory iteration" model will significantly enhance the model's logical reasoning limits and tool invocation accuracy when facing unknown complex tasks.

Currently, this MiniMax M2.7 model, equipped with self-evolution genes, has been fully launched on the MiniMax Agent platform and open platform. As large models begin to deeply participate in their own "growth" process, the ceiling of AI may be raised once again.

Meanwhile, the AI computing power and application market are also seeing frequent developments. LuChen Technology announced the completion of a Series B financing round worth hundreds of millions of yuan, with its overseas revenue share soaring to 79%; meanwhile, due to a surge in call volumes, some of Alibaba Cloud's AI computing power products have reportedly seen price increases. Amid the interplay of technological iteration and market fluctuations, the AI track in 2026 is becoming increasingly urgent and full of variables.

İlgili Sorular

QWhat is the name of the new model released by MiniMax that is capable of deep self-iteration?

AThe new model is called MiniMax M2.7.

QWhat is the core breakthrough of the MiniMax M2.7 model according to the article?

AIts core breakthrough is its strong autonomous construction capability, allowing it to build complex Agent Harness and complete highly complex productivity tasks independently.

QWhat specific abilities does the M2.7 model use to complete complex tasks?

AIt uses abilities such as Agent Teams (agent collaboration), complex Skills, and Tool Search tool to complete tasks.

QOn which platforms has the MiniMax M2.7 model been fully launched?

AIt has been fully launched on the MiniMax Agent platform and the open platform.

QBesides the MiniMax announcement, what other AI market dynamics are mentioned in the article?

AThe article mentions that LuChen Technology completed a Series B financing of hundreds of millions of yuan, and Alibaba Cloud increased prices for some AI computing products due to a surge in usage.

İlgili Okumalar

The Bitcoin 4-Year Cycle Has Never Disappeared, It Just 'Changed Shape'

Has Bitcoin's 4-Year Cycle Disappeared? The article argues that the fundamental 4-year cycle driven by halving events remains intact but its manifestation has changed due to a shift in the dominant buyer profile. The cycle peaked in October 2025, aligning with historical post-halving timelines. However, classic on-chain top indicators (MVRV, Pi Cycle, NUPL, etc.) failed to signal the peak. This is because these metrics are designed to detect retail FOMO, which was largely absent in Bitcoin this cycle. Retail capital was systematically drained by memecoins and high-FDV, low-float VC token launches before it could fuel a parabolic Bitcoin finale. Instead, institutional buyers (spot Bitcoin ETFs, corporate treasuries like MicroStrategy) became the marginal buyers. Their steady, programmatic accumulation created a strong but less volatile uptrend that didn't trigger traditional euphoria signals. The cycle's structure persists: a post-halving supply shock, a peak roughly 480-550 days later, and a bottom approximately one year after the peak (projected for Q3/Q4 2026). Bear market drawdowns are becoming shallower (projected 50-65% vs. past 78-86%) as institutional ownership matures. The article concludes that the cycle's timing remains the most reliable signal, while the tools to identify its phases must evolve with the market's structure.

marsbit6 dk önce

The Bitcoin 4-Year Cycle Has Never Disappeared, It Just 'Changed Shape'

marsbit6 dk önce

AI Agent Completely Transforms Web3 Gaming: From the Rugpull Bakery Bot Controversy to the New 2026 Agent Paradigm

This article explores how the AI Agent paradigm is fundamentally transforming Web3 gaming, moving from a disruptive force to a core, legitimized element. It begins with the controversy in the competitive baking game Rugpull Bakery, where automated scripts caused fairness issues. Instead of banning them, the developers integrated AI Agents into the official gameplay by providing technical documentation (skill.md, agent.json), marking a shift towards "Agentic Gaming." The piece outlines three primary implementation models for AI Agents in Web3 games by 2026: 1. **Autonomous Competitors & Economic Entities:** AI Agents act as independent players with unique strategies. Examples include TEN Protocol's poker agents, AI Arena's trainable NFT fighters, and Satoshi Strike Force's "Digital Athletes" trained on player data. The Somnia blockchain is highlighted as a dedicated "Agentic L1" infrastructure supporting this model at scale. 2. **Modular Infrastructure & Programmable Environments:** This model, exemplified by EVE Frontier, allows AI Agents to program game world logic itself. Using "Smart Assemblies" (e.g., Smart Turrets, Smart Gates), Agents can modify shared economic and physical rules on-chain, creating dynamic, player/AI-built worlds. The ERC-8183 standard further enables these automated entities to hire other AI services for complex tasks. 3. **Hybrid Companions & Dynamic Adaptation:** Here, AI serves as a collaborative partner. In Parallel Colony, highly autonomous AI Avatars work alongside human players who provide high-level guidance. Illuvium plans to use AI to make NPCs dynamic and responsive, creating personalized, emergent narratives for each player. The conclusion posits that Web3 gaming has reached a "post-human" inflection point. Blockchains' transparency and programmability, combined with new standards and infrastructure like Somnia, make integrating and governing AI Agents not just viable but essential. The future lies in a symbiotic digital order where players transition from manual laborers to commanders and partners of algorithmic intelligence.

marsbit22 dk önce

AI Agent Completely Transforms Web3 Gaming: From the Rugpull Bakery Bot Controversy to the New 2026 Agent Paradigm

marsbit22 dk önce

Saylor Bought 1,550 BTC, But This Might Be Strategy's Worst Trade Recently

Saylor purchased 1,550 BTC, but this may be one of MicroStrategy's worst recent trades for its MSTR shareholders. The article explains that for MSTR to increase its Bitcoin-per-share (BPS) metric, it must issue new shares at an mNAV (market cap to net asset value) ratio above a "breakeven" level, calculated here as 1.30. In this transaction, MSTR raised $181 million via an ATM offering and used only $101.3 million to buy Bitcoin, with the rest reportedly added to its dollar reserves. Crucially, the author argues the shares were likely sold below the 1.30 mNAV threshold. This combination—selling below breakeven mNAV and not deploying all proceeds into Bitcoin—resulted in a slight dilution of BPS, estimated at a 0.19% decrease. The trade is framed as a sacrifice of MSTR shareholder value (BPS) to extend the dollar reserve runway for its Sustainable Treasury Reserve Coin (STRC) subsidiary from 6.3 to 7 months. The author concludes that Strategy is gambling that this move will improve market sentiment for STRC and MSTR, allowing future value-accretive fundraising. If it fails, the company may face the difficult choice of further diluting MSTR or delaying STRC dividends.

marsbit26 dk önce

Saylor Bought 1,550 BTC, But This Might Be Strategy's Worst Trade Recently

marsbit26 dk önce

Saylor's Purchase of 1550 Bitcoin Is a Bad Trade

**Title: Saylor's Purchase of 1,550 Bitcoins Was a Bad Trade** The article critically analyzes Strategy's recent move of selling 32 bitcoins followed by a much larger purchase of 1,550 bitcoins. While appearing bullish, the author argues this trade is detrimental to MSTR shareholders. The core argument revolves around the concept of "breakeven modified Net Asset Value (mNAV)," a key metric for Strategy. To increase Bitcoin per share (BPS) for MSTR holders, Strategy must issue new shares at a premium high enough that the funds raised can buy more bitcoin than the bitcoin backing each existing share. Currently, this breakeven mNAV is estimated at 1.30. The recent trade failed on two counts: 1. The shares for the $181 million raise were issued at an mNAV *below* the 1.30 breakeven point. Selling "cheap" shares to buy bitcoin actually *reduces* BPS. 2. Only $101.3 million of the raised funds were used to buy bitcoin; the rest went to boost the company's dollar reserves. The breakeven mNAV calculation assumes *100%* of proceeds are used for bitcoin purchases. Diverting funds, even if mNAV were high, dilutes BPS. The result is an estimated 0.19% decrease in Bitcoin per share for MSTR holders. In exchange, Strategy merely extended its operational runway for its dollar reserves from ~6.3 months to 7 months. The author interprets this as Strategy prioritizing the survival and development of its STRC business over its stated core goal of increasing MSTR's BPS. This constitutes a gamble: if sacrificing MSTR value leads to improved market sentiment and a recovery in STRC's price (and thus mNAV), the whole system could work. If not, Strategy may be forced into a cycle of further diluting MSTR to stay afloat, potentially leading to deferred STRC dividends or corporate decline. The article concludes with a hope for price recovery for Bitcoin, MSTR, and STRC.

Foresight News33 dk önce

Saylor's Purchase of 1550 Bitcoin Is a Bad Trade

Foresight News33 dk önce

The AI Bear Market Lasting Two Days Is Over; Why Did Funds Buy Back Storage Stocks First?

After a severe two-day selloff in early June that erased over $1 trillion from U.S. chip stock market value, capital is flowing back first to the memory sector. The correction was not driven by a collapse in AI demand but rather a market reassessment of high expectations. Stocks like Broadcom faced selling pressure despite strong AI revenue guidance, signaling a shift in focus from who has an "AI story" to who can most rapidly translate AI demand into verifiable profits and earnings per share (EPS). Memory companies, such as Micron and SK Hynix, are leading the recovery because their EPS growth is more immediately verifiable. The AI server boom directly increases demand for high-bandwidth memory (HBM) and high-capacity server DRAM, tightening supply and driving up contract prices for conventional DRAM and NAND Flash. This price increase, coupled with a shift to higher-margin products, flows directly into near-term revenue and profitability, as evidenced in recent earnings reports. In contrast, other AI semiconductor segments like GPUs, ASICs, and optical modules, while central to the long-term AI infrastructure story, face longer and less certain paths to EPS validation. Their growth depends more on future product cycles, customer adoption timelines, and capital expenditure plans. The rebound in memory stocks highlights a market preference for assets with shorter, more transparent EPS conversion cycles following the recent de-risking phase. However, this does not negate the potential of other AI hardware segments should they provide clearer near-term order visibility. The episode has raised the validation bar for all AI-related investments.

marsbit33 dk önce

The AI Bear Market Lasting Two Days Is Over; Why Did Funds Buy Back Storage Stocks First?

marsbit33 dk önce

İşlemler

Spot

Futures