Over the past week, the AI market’s main battleground has shifted from “performance” to “price.” As OpenAI, Meta, and SpaceX (xAI) released new models in quick succession, capability leveled up across the board while challengers came in undercutting the leaders by more than half. The arena is moving from consumer to the enterprise (B2B) work market as well. It is the biggest shift in the AI landscape heading into this week’s Big Tech earnings season. 📊

TL;DR

  • From July 7–9, OpenAI (GPT-5.6), xAI (Grok 4.5), and Meta (Muse Spark 1.1) unveiled new models in a single week.
  • Challengers led with “half-price” tiers of $1–2 per million input tokens, and as performance leveled off, price became the key variable.
  • The contest is moving from consumer use to the enterprise (B2B) work and coding-agent market.

📅 A Week of New Models — What Happened

Over three days from July 7 to 9, three major AI firms shipped new models back to back. First, OpenAI globally launched its GPT-5.6 series. It split the lineup into the top-performing “Sol,” the balanced “Terra,” and the low-cost, high-speed “Luna” — a structure meant to let users pick by task difficulty and cost.

That same week, Elon Musk’s xAI (SpaceX) fully unveiled “Grok 4.5” on July 9, specialized for coding and agentic tasks. Meta also introduced “Muse Spark 1.1” the same day (local time) — an agent-type model that handles multi-step work such as web search, document analysis, and code writing on its own. All three share a focus on doing actual “work” rather than simple conversation.

💸 The Real Story Is ‘Price’ — the Half-Price Model Race

The core of this release rush was price, not performance. As challengers arrived with API fees less than half of the front-runners’, an “AI price war” effectively began.

  • Grok 4.5: $2 input / $6 output per million tokens
  • Meta Muse Spark 1.1: $1.25 input / $4.25 output per million tokens
  • GPT-5.6 Luna: $1 input / $6 output per million tokens

The gap becomes stark against premium models. Some top-tier models run $5–10 input and $25–50 output per million tokens — roughly five times the challengers’ low-cost tiers. Now that the performance gap has narrowed, how cheaply you can get the same job done has become the decisive variable for adoption.

🏆 From “Big Three” to “Two Leaders, Three Chasers”? — A Reshuffling Field

The market structure is shifting too. The Korea Economic Daily framed the Anthropic–OpenAI–Google “big three” as moving toward a “two leaders, three chasers” setup: Anthropic and OpenAI lead on model performance, while Google, Meta, and xAI — armed with large infrastructure — give chase. That said, it is one outlet’s analytical frame, so take it as a reference point.

The chasers’ weapons are infrastructure and data. xAI is the prime example: on June 16 it announced it would acquire the maker of the coding tool “Cursor” in an all-stock deal worth roughly $60 billion (expected to close in Q3). Grok 4.5 is reported to have trained on Cursor’s real developer session data to boost its coding performance.

🏢 The Battleground Is the Enterprise (B2B) Work Market

Beneath the price competition lies a shared goal: to lock down the enterprise market first. Each company’s strategy, however, differs slightly.

  • Anthropic: Leaning on its coding agent (Claude Code), it is doubling down on B2B such as development and workflow automation. Per some reports, its enterprise customers spending over $100 million a year have surpassed 1,000 (the figure is a media estimate).
  • OpenAI: Shifting its center of gravity from consumer to enterprise, it targets the business market with its previously unveiled work agent “ChatGPT Work” and coding tool “Codex.”
  • Google: It ties Gemini into its own services — search, mail, documents — and adds a relatively affordable developer plan (around $100 a month). It is a full-stack strategy aimed at consumers and enterprises at once.

On the enterprise side, rather than sticking to one vendor’s model, a “multi-model” approach that mixes several models to optimize cost and efficiency is taking hold. AI firms differentiating on performance, price, and focus area also dovetails with this demand.

🇰🇷 What It Means for Korea

This trend has immediate implications at home. First, as low-cost models proliferate, the barrier and running cost for Korean firms to adopt AI drop. Multi-model adoption — swapping models to fit the situation — could accelerate too. Conversely, domestic AI players like Naver and Kakao face greater competitive pressure on both price and performance. And the hotter the model race, the more likely demand holds for the high-performance memory (HBM) and foundry capacity used in training and inference — a potential opportunity for Korea’s chip industry.

🧭 The Takeaway

The key point this week is that the axis of AI competition has moved from performance to cost and work output. The models that poured out over the week competed less on bragging rights and more on “how cheaply, and how much actual work they can take over.” For consumers and businesses it means wider choice and lower costs, but for AI firms it comes with the homework of profitability pressure. Three things to watch ahead: how the price war shakes each company’s revenue structure, whether performance, price, or integration convenience decides the enterprise market, and whether this week’s Big Tech earnings keep AI investment (capex) plans robust enough to sustain the fight.

※ This article is for informational purposes only and is not investment advice.

Sources