Elon Musk's SpaceXAI has launched Grok 4.5, a cost-effective AI model targeting developers and engineers, significantly underpricing competitors.
Summary
- SpaceXAI unveiled Grok 4.5 on July 8, offering competitive pricing at $2 per million input tokens and $6 for output, undercutting rival models.
- Elon Musk stated on X that Grok 4.5 is similar to Anthropic's Opus 4.7 but emphasizes cost and speed over performance.
- Grok 4.5 is currently unavailable in the EU, but SpaceXAI plans to release it there by mid-July.
On July 8, Elon Musk's SpaceXAI launched Grok 4.5, its first model release since merging with SpaceX-xAI in February. This launch occurs amid SpaceX's ongoing $60 billion acquisition of Cursor. The model is designed for software developers, engineers, and various "knowledge workers" including legal and finance professionals.
Instead of positioning itself as the top performer, SpaceXAI highlights Grok 4.5's affordability compared to competitors. Its pricing structure is significantly lower than that of rival models, with Anthropic's Claude Opus 4.8 charging $5 for input and $25 for output, while OpenAI's GPT 5.6 Sol costs $5 for input and $30 for output.
Musk’s announcement on X described Grok 4.5 as "roughly comparable to Opus 4.7, but much quicker," focusing on speed and cost rather than peak performance. He cited engineers at Tesla and SpaceX as examples of practical uses for the model.
Benchmark Performance
Upon its introduction, SpaceXAI shared four benchmark results highlighting a varied performance profile. In the DeepSWE 1.1 benchmark, which tests the ability to fix software bugs, Grok 4.5 scored 53%, trailing behind Claude Opus 4.8 at 59% and GPT 5.5 at 67%. The top scorer was Claude Fable 5 at 70%.
In SWE Bench Pro, which assesses software engineering tasks, Grok 4.5 achieved 64.7%, surpassing GPT 5.5's 58.6%, but still behind Opus 4.8 at 69.2% and Fable 5 at 80.4%. Notably, SpaceXAI chose to compare Grok 4.5 with GPT 5.5 instead of the newly launched GPT 5.6.
Grok 4.5 was developed with Cursor AI, utilizing a substantial amount of Nvidia GB300 GPUs within the Colossus supercomputer, which has over 200,000 GPUs. Although Grok 4.5 is competitive, it does not lead in performance metrics.
Cost Efficiency
The primary advantage of Grok 4.5 lies in its cost efficiency. In SWE Bench Pro tasks, it used an average of 15,954 output tokens per task, significantly lower than Opus 4.8's 67,020 tokens, indicating a 4.2x efficiency advantage.
This efficiency can lead to considerable savings for teams using AI, especially with Grok 4.5's already reduced token costs. While its performance in benchmarks may not be the highest, its cost-effectiveness allows for more iterations at a lower expense.
Operating at a speed of 80 tokens per second, Grok 4.5 is classified as a fast model. It was trained using data from developer sessions at Cursor, including debugging logs and code changes, addressing past criticisms of xAI’s training methods.
For developers with high-volume coding needs, Grok 4.5 offers capabilities similar to Opus 4.7 at a 60% lower cost per input token. However, for those seeking the latest advancements, Claude Fable 5 remains the top choice in the evaluated categories. Initial tests using Grok on Hermes showed disappointing results for creative writing, while basic coding tasks yielded adequate performance.
The model is available via API, on Hermes, and through Grok build, featuring a context window of half a million tokens (nearly 400,000 words). European users will have to wait, as SpaceXAI plans to release Grok 4.5 in the EU by mid-July.