Musk wants to wage a price war—if Grok 4.5 can’t make money, it will still go after Claude. Is this good news for SpaceX?

TL;DR

· SpaceXAI released Grok 4.5 on July 8 and integrated it with Cursor, Grok Build, and the API.

· It may not comprehensively outperform OpenAI or Anthropic, but low pricing is testing whether premium high-end API pricing can hold.

· Related tickers: SpaceX/SPCX exposure, OpenAI and Anthropic-related proxies, AI developer toolchain.

SpaceXAI released Grok 4.5 on July 8 and integrated it with Grok Build, Cursor, and the SpaceXAI/xAI API. Official documentation shows that Grok 4.5 is aimed at coding, agent tasks, and knowledge work, with pricing of $2 per million input tokens and $6 per million output tokens.

Grok 4.5 is positioned to take on Claude Opus 4.8, which costs $5 for input and $25 for output.

This release has drawn attention from developers and investors, and it’s not just because there’s another cutting-edge model in the market. The more direct question is: if a near-top coding and agent model can complete similar tasks at lower cost, how long can OpenAI and Anthropic’s premium pricing for high-end APIs last?

Here, tokens can be understood as units of text and code snippets used for AI billing. An agent is an AI assistant that can break down steps on its own, call tools, and iteratively revise results. For enterprises, where a model ranks on leaderboards matters, but the actual cost of writing code, fixing a bug, and running an automation workflow is becoming just as important.

Pulling competition back to task cost with low prices

The first layer of impact from Grok 4.5 comes from pricing. Official documentation and the Cursor blog confirm that its API costs $2 per million input tokens and $6 per million output tokens, lower than the common pricing range of some high-end models.

But this isn’t just “being cheap.” The accounting for AI coding tools usually isn’t calculated per single question—it’s calculated per task. Complex tasks often involve multiple rounds of actions, such as reading code, writing code, running tests, fixing errors, and then resubmitting results. The more capable the model is at planning and the less it takes detours, the fewer tokens are actually consumed.

So, unit task cost is closer to the real procurement logic enterprises use than the price of a single token. If Grok 4.5 can maintain a high completion rate in real engineering tasks, the low unit price will be further amplified into a lower total cost per task.

Artificial Analysis’ evaluation provided support for this narrative. The firm said Grok 4.5 ranks 4th in the Intelligence Index, behind Fable 5, GPT-5.5, and Opus 4.8, but sits on the effective performance-and-cost frontier and shows lower coding agent costs.

This also explains why market discussion is concentrating on the “price anchor.” In the past, competition among frontier models was often framed as who is the strongest, and users were willing to pay for the strongest capabilities. In coding and agent scenarios, if a model close to the first tier is cheap enough, enterprises start comparing which one is more cost-effective.

Cursor provides an entry point for low-priced models

If Grok 4.5 is only a low-cost API, it’s more like a price war. Once integrated into Cursor, the competition becomes one among the model, distribution, and developer workflows.

Cursor is a widely used AI coding tool for developers. Its position isn’t a typical call interface; it’s the workspace where developers write code, modify code, and debug projects every day. After a model enters this entry point, it’s easier to be used frequently and easier to become the default choice.

This is also the strategic implication of SpaceX acquiring Cursor’s parent company, Anysphere. SpaceX plans to acquire Anysphere in an all-stock deal valued at $60 billion, with the transaction expected to be completed in the third quarter of 2026. This framing needs to preserve boundaries: it’s more appropriate to interpret it as an expected equity tie-up, not already-realized financial synergies.

From a product logic perspective, Grok 4.5 has already demonstrated the problem that this integration line is trying to solve. Model companies are no longer just selling APIs—they’re trying to embed models into the development workflows that developers use most often.

A data flywheel could form here. Developers use AI in Cursor to write code, which generates real task data. If these data can be used compliantly to improve engineering capabilities, the model will understand code contexts better and, in turn, increase Cursor’s stickiness.

But the flywheel is still a potential outcome, not an already-established moat. It needs real usage volume, retention rates, and task completion rates to validate. A single release can only indicate that SpaceXAI is tying the model to an entry point; it can’t prove that developers have already migrated at scale.

OpenAI and Anthropic face a premium pricing test

For OpenAI and Anthropic, Grok 4.5 may not cause immediate customer churn in the short term. When enterprises choose models, they look not only at price, but also at safety, stability, context capabilities, tool ecosystems, compliance support, and service responsiveness.

What’s being tested is the price elasticity of high-end APIs. In the past, top models could rely on their strongest capabilities to sustain higher pricing, especially in complex reasoning, code generation, and enterprise automation scenarios. Now, if models like Grok 4.5 are close enough to top-tier capability across most engineering tasks, customers will begin to break down requirements.

A more likely change is layered usage. Enterprises may still assign their most difficult, most sensitive, and most complex tasks to the strongest models, but move large volumes of daily coding, testing, documentation, and office automation tasks to lower-cost models.

This won’t immediately eliminate the value of high-end models, but it could compress the space where they cover the long tail of tasks. For OpenAI and Anthropic, the issue may not be losing the most core use cases; instead, mid-to-lower complexity tasks that used to be covered with premium pricing could start being taken by cheaper models.

Musk and the media describe Grok 4.5 as an “Opus-level” model, and Axios also mentioned that it is aiming to compete on top capabilities, while pointing out that it doesn’t exceed all of OpenAI and Anthropic’s biggest or newest models. This qualification matters. The impact of Grok 4.5 isn’t that it has already comprehensively surpassed them; it’s that it makes customers recalculate the balance between “strong enough” and “cheap enough.”

For investors, this affects the valuation narrative. The high valuations of OpenAI and Anthropic rely on frontier model capabilities, enterprise subscriptions, and API revenue expansion. If high-end API pricing is forced downward, the market will focus more on revenue quality, gross margins, and customer stickiness—not just model capability iteration.

Adoption rate determines whether the price anchor loosens

The assessment Grok 4.5 supports right now is that it has delivered a low-price, high-performance narrative in coding and agent scenarios, shifting the competition focus from model capability leaderboards to unit task cost. It still can’t be said that SpaceXAI has already reshaped the AI coding market.

The tougher variable is inside Cursor. Whether developers set Grok 4.5 as the default model, whether enterprises are willing to move high-frequency tasks there, and whether task completion rates can be stably replicated in real projects—all of that matters more than the buzz on release day.

The response speed of competitors will also determine how far the impact spreads. If OpenAI and Anthropic roll out lower-priced tiers, adjust cache pricing, or strengthen enterprise security and proprietary data advantages, Grok 4.5’s cost advantage could be partially absorbed. Conversely, if competitor pricing holds strong while Cursor usage continues to expand, the price anchor on high-end APIs will loosen even faster.

This release hasn’t delivered a winner-or-loser conclusion yet. It’s more like pushing competition among AI coding models into a new stage: frontier capabilities remain important, but investors need to keep an eye on unit task cost, distribution entry points, and real adoption rates at the same time. Grok 4.5 puts this arithmetic problem on the table first.

Click to learn about positions at Lyt Blocks Beats

Welcome to join the official Lyt Blocks Beats community:

Telegram subscription group: https://t.me/theblockbeats

Telegram discussion group: https://t.me/BlockBeats_App

Twitter official account: https://twitter.com/BlockBeatsAsia