AI Giants' Intense 72-Hour Releases: Model Race Accelerates, Industry Landscape Shifts

Industry Context

Over the past three days, the world’s leading AI labs have fallen into an unusually dense release schedule. OpenAI, Meta, SpaceXAI (formerly xAI), and Anthropic have successively rolled out new models and features, with major updates arriving almost every few hours. This “arms race” is no accident—since the second quarter of 2026, open-source models such as Muse Spark 1.1 have begun challenging closed-source ones, with Meta charging for its model for the first time; OpenAI is shoring up its moat through the GPT-5.6 series (flagship Sol, everyday Terra, low-cost Luna); SpaceXAI released Grok 4.5 after acquiring Cursor and publicly signaled goodwill toward Anthropic; Anthropic strengthened the user experience with its “Reflect” feature. Behind all this lies a critical inflection point for the AI industry, shifting from model capability competition to commercialization and ecosystem competition.

Market Impact

For enterprise customers, the range of model choices has surged and cost structures are diverging. OpenAI’s Luna model focuses on cost savings, Meta’s Muse Spark 1.1 hits the market at “extremely low prices,” and SpaceXAI’s Grok 4.5 positions itself as “Opus-level.” Companies can choose flagship or economy models based on budget and use case, but they also face the risk of vendor lock-in. For investors, the dense release cadence has intensified scrutiny of AI companies’ profitability—Meta’s first-ever charge signals its exploration of a commercialization path; OpenAI has strengthened enterprise value by integrating Codex and launching ChatGPT Work. Perplexity was reportedly developing a programming tool called “Teammate,” further stretching valuation logic.

Competitive Landscape

OpenAI: Released the GPT-5.6 family, integrated Codex into the desktop app, launched the enterprise agent “ChatGPT Work,” while canceling the Atlas browser and scaling back “sideline businesses.” CEO Sam Altman is attempting to cover the full spectrum from consumers to enterprises with its product matrix.

Meta: Through the superintelligence division led by Alexandr Wang, it has churned out continuous releases—Muse Image sparked privacy controversy (default public Instagram accounts) and Muse Spark 1.1 was priced for the first time. Zuckerberg publicly criticized competitors’ pricing as “extreme,” aiming to occupy the high ground of cost-effectiveness.SpaceXAI: Musk publicly admitted "previous misjudgment of Anthropic" and revealed that SpaceXAI pays Anthropic $1.25 billion per month for computing power. The release of Grok 4.5, combined with the acquisition of Cursor, suggests an integration of programming capabilities.

Anthropic: Launched the "Reflect" feature, allowing users to see their own usage habit analysis, differentiating itself. It also received computing power support from SpaceXAI, shifting the relationship from competition to cooperation.

Perplexity: Quietly developing a programming tool called "Teammate", aiming to challenge Cursor and OpenAI Codex, becoming a new variable.

Enterprise Implications

Enterprises should focus on the following three points: 1. Model selection strategy: No longer solely based on the "strongest model" theory. Flagship models (such as Sol, Grok 4.5) are suitable for high-precision scenarios, while economical models (such as Luna, Muse Spark 1.1) are suitable for large-scale deployment. Enterprises need to establish a cost-performance evaluation framework. 2. Ecosystem lock-in risk: OpenAI integrates Codex into desktop apps, Meta defaults to making data public. Enterprises need to assess data sovereignty and migration costs. 3. Agentification trend: OpenAI's ChatGPT Work and Perplexity's Teammate indicate that AI agents are evolving from tools to workflow infrastructure. Enterprises should begin piloting ROI of agents in scenarios such as customer service, programming, and data analysis.

Outlook

Within 12 months: The frequency of model releases will remain high, but differentiation will shift from parameters to pricing, integration, and agent capabilities. Meta's low-price strategy may force other vendors to follow suit, and inference costs will drop rapidly.

Within 24 months: The market may see a wave of consolidation—OpenAI scaling back non-core products (such as Atlas), SpaceXAI and Anthropic forming an alliance, and new players like Perplexity entering through niche tools. Enterprise AI applications will shift from "using models" to "deploying agents."

3 years: The boundary between open source and closed source will blur. Meta's monetization attempts and OpenAI's closed-source approach will coexist, but user conversion costs will determine the winner. AI regulation (such as the EU AI Act) may impose restrictions on default public training data (e.g., Muse Image), affecting business models.