TL;DR
· SpaceXAI released Grok 4.5 on July 8 and integrated it with Cursor, Grok Build, and the API.
· It may not comprehensively lead OpenAI and Anthropic, but low pricing is testing whether there’s still a premium for high-end APIs.
· Related plays: SpaceX/SPCX exposure, OpenAI and Anthropic-related proxies, AI developer toolchains.
SpaceXAI released Grok 4.5 on July 8 and integrated it with Grok Build, Cursor, and the SpaceXAI/xAI API. The official documentation shows that Grok 4.5 is designed for coding, agent tasks, and knowledge work, priced at $2 per million input tokens and $6 per million output tokens.
Grok 4.5 is aiming to match Claude Opus 4.8, which is priced at $5 for inputs and $25 for outputs.
This release has attracted attention from developers and investors—not just because the market has another cutting-edge model. The more direct question is: if a near-top-tier coding and agent model can complete similar tasks at lower cost, how long can OpenAI and Anthropic’s high-end API premiums hold?
Here, “tokens” can be understood as units of text and code snippets used in 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 a piece of code, fixing a bug, and running an automation workflow is becoming just as important.
Low pricing brings competition back to task cost
The first wave of impact from Grok 4.5 comes from price. The official documentation and the Cursor blog confirm that its API is $2 per million input tokens and $6 per million output tokens—below the common price range of some high-end models.
But this isn’t just “cheap.” For AI coding tools, billing usually isn’t based on asking one question; it’s based on completing a task. Complex tasks often involve multiple rounds of actions, such as reading code, writing code, running tests, fixing errors, and resubmitting results. The more the model plans and the fewer detours it takes, the fewer tokens it consumes in practice.
So, unit task cost is closer to how enterprises truly procure 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 could be amplified into even lower total task costs.
Artificial Analysis’s evaluation provides 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 has centered on the “price anchor.” In the past, competition among frontier models was often framed as who is strongest, and users would pay for the strongest capabilities. But in coding and agent scenarios, if a model close to the first tier is cheap enough, enterprises begin comparing which one is more cost-effective.
Cursor provides an entry point for low-priced models
If Grok 4.5 is merely a low-priced API, then it’s more like a price war. Once integrated with Cursor, the competition becomes one among models, distribution, and developer workflows.
Cursor is a commonly used AI coding tool for developers. Its role isn’t a typical calling interface—it’s the workspace where developers write code, modify code, and debug projects every day. After the model enters this entry point, it becomes easier to use frequently and easier to become the default choice.
This is also the strategic meaning behind SpaceX’s acquisition of 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 keep its boundaries—it’s better understood as expected equity tie-ups rather than financial synergies that have already been realized.
From a product perspective, Grok 4.5 has already shown the integration line it is trying to solve. Model companies are no longer just selling APIs; they are trying to place the model into the workflows developers use most often.
A data flywheel could form here. When developers use AI in Cursor to write code, it generates real task data. If this data can be used compliantly to improve engineering capabilities, the model will understand coding contexts better—then, in turn, increasing Cursor’s stickiness.
But the flywheel is still a potential outcome, not a moat that has already been established. It needs to be validated by real usage volume, retention rate, and task completion rate. A single release can only show that SpaceXAI is tying the model to an entry point; it cannot yet prove that developers have migrated at scale.
OpenAI and Anthropic face a premium test
For OpenAI and Anthropic, Grok 4.5 may not cause core customer churn in the short term. When enterprises choose models, they don’t look at price alone—they also consider security, stability, context capabilities, tool ecosystems, compliance support, and service responsiveness.
What is being tested is the price elasticity of high-end APIs. In the past, top models could rely on their strongest capabilities to sustain higher prices, 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 splitting requirements.
One change that’s more likely to happen is tiered usage. Enterprises may still assign their hardest, most sensitive, and most complex tasks to the strongest models, but shift large volumes of everyday coding, testing, documentation, and office automation tasks to lower-cost models.
This won’t immediately wipe out the value of high-end models, but it could compress their room to cover long-tail tasks. For OpenAI and Anthropic, the issue may not be losing the most core scenarios; it may be that mid-to-low complexity tasks that could previously be covered at high prices start getting taken by cheaper models.
Musk and the media have described Grok 4.5 as an “Opus-level” model. Axios also mentioned that it is aiming to compete for top-tier capabilities, while pointing out that it does not exceed all of OpenAI and Anthropic’s biggest or latest models. This qualification is important. Grok 4.5’s impact isn’t that it has already surpassed them comprehensively; it’s that it makes customers reassess the balance between “strong enough” and “cheap enough.”
For investors, this will affect valuation narratives. OpenAI and Anthropic’s high valuations depend on frontier model capabilities, enterprise subscriptions, and expansion of API revenue. If high-end API pricing is forced downward, the market may focus more on revenue quality, gross margin, and customer stickiness—not just on model capability iterations.
Adoption rate determines whether the price anchor loosens
What Grok 4.5 can currently support is the judgment that it has delivered a low-price, high-performance narrative in coding and agent scenarios, shifting the competitive focus from model capability leaderboards to unit task cost. It also still can’t be said that SpaceXAI has already reshaped the AI coding market.
The tougher variable lies within Cursor. Whether developers set Grok 4.5 as the default model, whether enterprises are willing to migrate high-frequency tasks over, and whether task completion rates can be stably reproduced in real projects—these matter more than the buzz on release day.
The speed at which competitors counter will also determine the scope of the impact. If OpenAI and Anthropic roll out lower-priced tiers, adjust cache prices, or strengthen enterprise security and proprietary data advantages, Grok 4.5’s cost advantage could be partially absorbed. Conversely, if competitor pricing remains strong while Cursor usage continues to expand, the price anchor for high-end APIs will loosen even faster.
This release has not yet provided a verdict. It’s more like pushing competition among AI coding models into a new stage: frontier capabilities remain important, but investors need to watch unit task cost, distribution entry points, and real adoption rates at the same time. Grok 4.5 puts this arithmetic question on the table first.
Click to learn about BlockBeats Lawtide recruiting positions
Welcome to join the official BlockBeats Lawtide community:
Telegram subscription group: https://t.me/theblockbeats
Telegram discussion group: https://t.me/BlockBeats_App
Twitter official account: https://twitter.com/BlockBeatsAsia
