# SpaceXAI Releases Grok 4.5: Opus-Class Coding Model at $2 per Million Tokens

> The new release from SpaceXAI delivers frontier-level performance for coding and agentic tasks through joint training with Cursor, combining high intelligence with notable advantages in speed, token efficiency, and pricing.

*Published 2026-07-09 · By Marcus Vance*

Grok 4.5 is SpaceXAI's smartest model built for coding, agentic tasks, and knowledge work.

SpaceXAI has introduced Grok 4.5 as its latest offering in the frontier models space. The announcement highlights the model's focus on coding and agent intelligence, setting it apart from general purpose models that are trained on broad internet data without specific focus on developer workflows. Trained jointly with Cursor, it incorporates data from developer interactions to enhance its capabilities in software engineering tasks. This collaboration represents a strategic move to leverage specialized data for improved performance in real world scenarios where developers interact with codebases of varying complexity. The model is described as the smartest from the company and the first specifically built for more than software engineering according to the joint announcement with Cursor. This indicates a broadening of scope to include agentic tasks and knowledge work in addition to pure coding activities. Such a focus could allow the model to assist in tasks like debugging complex systems, generating agent behaviors, and performing knowledge retrieval within large code repositories. The emphasis on these areas shows a commitment to practical utility in professional settings.

The development of Grok 4.5 builds on a foundation model with 1.5 trillion parameters known as V9. Additional training from Cursor data and reinforcement learning on difficult multi-step problems have been applied to refine its abilities. Such training methods allow the model to handle complex agentic tasks that require sequential decision making and problem solving over multiple steps. The emphasis on these areas positions Grok 4.5 as a tool for knowledge work beyond simple code generation. By incorporating trillions of tokens from actual developer sessions the model learns patterns that reflect real usage rather than synthetic examples alone. This results in better handling of edge cases commonly encountered in large scale software projects. The supplemental training phase after the base V9 model ensures specialization without sacrificing the broad capabilities expected from frontier class systems.

## What are the performance characteristics of Grok 4.5?

Grok 4.5 operates at speeds of 80 tokens per second, which is described as fast-model speeds suitable for interactive use. It also demonstrates roughly 2 times greater token efficiency than leading models on engineering tasks. This efficiency translates to lower resource usage and faster response times in practical deployments. The combination of speed and efficiency makes it suitable for real-time agent applications where latency is critical. Users running long sessions benefit from reduced compute demands allowing more queries within the same budget. The design choices reflect a focus on production readiness rather than raw benchmark chasing alone. Engineering teams can integrate the model into continuous integration pipelines without concerns over excessive delays or costs.

On the SWE Bench Pro benchmark, the model shows significant improvements in token usage. It produces 4.2 times fewer output tokens on average compared to Opus 4.8. The average output for Grok 4.5 is 15,954 tokens versus 67,020 for the competing model. This reduction in token consumption directly impacts the cost and speed of operations for users running complex engineering workflows. Lower token counts also mean less context window pressure during extended interactions with agents. The efficiency advantage compounds over multiple agent turns making Grok 4.5 particularly effective for multi-step problem resolution. These metrics were measured under controlled conditions that simulate professional coding environments.

## How is Grok 4.5 priced and what is its availability?

The API pricing for Grok 4.5 is set at $2 per million input tokens and $6 per million output tokens. This pricing structure undercuts many rivals in the market for high intelligence models. The low cost combined with high performance makes it an attractive option for developers and companies looking to integrate advanced AI into their workflows without incurring high expenses. Organizations with high volume usage will see substantial savings compared to models with higher per token rates. The output price remains competitive even for verbose responses typical in code generation and agent planning. This economic model encourages experimentation and broader deployment across teams of varying sizes.

Grok 4.5 is the default model in Grok Build and is available across all Cursor plans. This wide availability ensures that users of these platforms can immediately access the model's capabilities. The integration into existing tools like Cursor allows for seamless adoption by software engineers who are already using such environments for their daily work. Teams using multiple plans can standardize on Grok 4.5 without additional configuration overhead. The default status in Grok Build further simplifies access for new users exploring agentic coding workflows. Availability across plans supports both individual developers and enterprise customers with different usage requirements.

Grok 4.5 Specifications and PricingMetricValueFoundation Model1.5T parameter V9Input Price$2 per million tokensOutput Price$6 per million tokensInference Speed80 TPSToken Efficiency2x greater than leading models on engineering tasksTraining DataTrillions of Cursor developer tokens

## What training approach was used for Grok 4.5?

The model was built on a 1.5T parameter V9 foundation model with supplemental Cursor training. This was followed by extensive RL on difficult multi-step problems. The use of trillions of tokens of Cursor data on developer interactions and codebases provides a rich dataset that reflects actual usage scenarios in software development. This data-driven approach helps the model learn patterns that are relevant to coding and agentic tasks specifically. The RL phase targets challenging scenarios that require planning and iteration. By focusing training resources on these areas the model gains robustness for real agent deployments. The joint effort with Cursor ensures the data distribution matches the target use cases closely.

- Built on 1.5T parameter V9 foundation model
- Incorporated trillions of tokens from Cursor developer data
- Applied extensive reinforcement learning on multi-step problems
- Optimized for coding and agentic performance at 80 TPS

## What are the market implications of Grok 4.5's release?

The release of Grok 4.5 at a competitive price point of $2 per million input tokens signals a shift in the economics of frontier model usage. Developers and enterprises can now access Opus-class intelligence without the premium costs associated with other providers. This could lead to broader adoption of advanced AI in coding tools and agent systems across various industries. The efficiency gains also mean that larger scale deployments become more feasible from a cost perspective. Smaller teams gain access to capabilities previously limited to well funded organizations. The market may see increased competition as other providers adjust their offerings in response to this pricing benchmark. Long term this could accelerate innovation in agent frameworks that rely on frequent model calls.

Stakeholders in the AI ecosystem, including tool providers like Cursor and competitors in the coding model space, will need to respond to this pricing and performance combination. SpaceXAI's focus on agent intelligence suggests a move toward more autonomous systems that can handle knowledge work. This has the potential to accelerate the development of AI agents that can perform complex tasks with less human intervention. Cursor users benefit from immediate access through existing plans while new entrants to agent development gain a cost effective entry point. The combination of speed and efficiency supports use cases in continuous development environments where response time matters. Overall the release contributes to a more accessible frontier model landscape.

## What reactions have been noted regarding Grok 4.5?

According to the announcement from SpaceXAI, the model received strong positive feedback from customers in the beta test program. This feedback prompted the decision to make Grok 4.5 available to the public. The model's design as faster, more token-efficient, and lower cost than typical Opus-class models has been highlighted as a key advantage. Beta participants reportedly found the performance suitable for demanding engineering workloads. The public rollout follows this validation phase to ensure reliability at scale. Reactions emphasize the practical benefits over pure capability increases seen in other recent releases.

> It is an Opus-class model, but faster, more token-efficient and lower cost.Elon Musk, CEO, SpaceXAI

## What can be expected next for Grok 4.5?

With the public release, users can expect Grok 4.5 to be integrated more widely into development environments and agent platforms. Future updates may build on the current foundation to further improve its capabilities in multi-step reasoning and code handling. The partnership with Cursor indicates ongoing collaboration that could lead to even more specialized versions tailored to specific developer needs. Continued data collection from usage will likely inform refinements in subsequent iterations. The model is positioned for iterative improvement based on real world agent deployments. SpaceXAI may expand availability to additional platforms as demand grows.

The overall trajectory for SpaceXAI appears to be toward models that prioritize practical applications in coding and agents while maintaining cost efficiency. This approach could influence the broader industry to focus on similar optimizations in their model releases. As more data from real world use becomes available, refinements to Grok 4.5 are likely to follow. Developers should monitor updates for enhancements in agent reliability and integration features. The current release establishes a baseline for cost effective frontier intelligence that subsequent versions can build upon. Industry observers will watch for how competitors position their offerings against this benchmark.

## Sources

1. [Today, we're launching Grok 4.5 , SpaceXAI's smartest model built to excel at coding, agentic tasks, and knowledge work. It's our strongest model ever and was trained alongside Cursor. Pricing is $2 per million input and $6 per million output with 4.2× fewer output tokens than Opus 4.8 on SWE Bench Pro.](https://x.ai/news/grok-4-5)
2. [Today we are releasing Grok 4.5 together with SpaceXAI , our most intelligent model and the first we've built for more than software engineering.](https://cursor.com/blog/grok-4-5)
3. [Based on strong positive feedback from customers in our beta test program, @SpaceXAI will make Grok 4.5 available to the public tomorrow. It is an Opus-class model, but faster, more token-efficient and lower cost.](https://x.com/elonmusk/status/2074740539874775163)

---
Source: https://aiintelreport.com/frontier-models/spacexai-releases-grok-4-5
Index: https://aiintelreport.com/llms.txt · Full text: https://aiintelreport.com/llms-full.txt