The AI agent market has gained a new undisputed leader. The analytical agency Artificial Analysis conducted an independent benchmark, AutomationBench-AA, and the results proved highly revealing. The Grok 4.5 model from SpaceXAI not only surpassed recognized giants — Claude Fable 5 and Claude Opus 4.8 — but did so with a massive lead in cost efficiency.
Victory on all fronts: performance and cost
In the test, which included 657 tasks across 40 simulated applications (Gmail, Slack, Salesforce, HubSpot, and others), Grok 4.5 achieved a 51.4% success rate. For comparison, Claude Fable 5 scored 48.6%, and Claude Opus 4.8 scored 48.5%. However, the main surprise lies in the price. Completing one task with Grok 4.5 costs only $0.34, while Fable 5 costs $1.35, and Opus 4.8 costs $1.46. The closest competitor in price, Gemini 3.5 Flash, costs $0.49 per task but shows significantly more modest results.
Efficiency as a key advantage
The secret to Grok 4.5's success lies in its V9 architecture with 1.5 trillion parameters. To complete one task, the model uses about 8,000 output tokens — roughly 25% of the resources consumed by Opus 4.8. The total token consumption per task is only 0.44 million — one of the lowest among all models. It is this efficiency, multiplied by the low token price, that creates the competitive advantage SpaceXAI representatives spoke about at launch.
Financial segment: undisputed dominance
Special attention should be paid to Grok 4.5's performance in the financial sector — the most challenging category of the test. The model achieved a 71% success rate, leaving behind Fable 5 (64%) and Opus 4.8 (62%). This is a critically important indicator for companies integrating AI agents into real financial systems. However, it is worth noting that Grok 4.5 violates rules more often than its competitors: an average of 0.63 violations per task, compared to 0.55 for Opus 4.8 and 0.46 for Gemini 3.5 Flash. In high-risk scenarios, this could become a serious drawback.
My analysis: Grok 4.5 demonstrates that the future of AI agents lies not just in "smart" models, but in cost-effective solutions. A fourfold price gap with comparable task quality makes this model extremely attractive for mass adoption in business processes. However, the higher rate of rule violations requires additional attention and tuning, especially in regulated sectors such as finance. The AI agent market is clearly entering a phase of price wars, and this is excellent news for end users.