Gemini 2.5 Pro is still king, but GPT-5-mini comes surprisingly close at 1/5th the cost

Assessing AI Model Performance: Gemini 2.5 Pro Versus GPT-5-mini — A Cost-Performance Breakdown

In the rapidly evolving landscape of language models, staying informed about performance benchmarks and cost-efficiency is essential for making strategic choices. Recently, I conducted an in-depth evaluation of popular AI models, focusing on Gemini 2.5 Pro and the newly available GPT-5-mini, to determine which offers the best value for various applications.

Why Gemini 2.5 Pro Continues to Lead in Performance

Throughout my testing, Gemini 2.5 Pro has demonstrated exceptional capabilities, maintaining its status as a top-tier model. Its standout metrics include:

• Median Score in SQL Query Generation: 0.967, the highest among tested models
• Success Rate: 88.76%, indicating reliable performance across diverse tasks
• Cost Efficiency: Priced at $1.25 per million tokens, delivering significant value

These results reinforce the previous perception that Google’s Gemini suite provides some of the most robust AI solutions on the market today.

Introducing an Unexpected Contender: GPT-5-mini

Despite initial disappointment with GPT-5’s performance at release, I was intrigued to test GPT-5-mini, expecting subpar results. Surprisingly, it closely rivals Gemini 2.5 Pro on core metrics at a fraction of the cost, presenting an intriguing alternative for large-scale or budget-conscious projects.

| Model | Median Score | Average Score | Success Rate | Cost (per million tokens) |
|———|—————-|—————-|————–|—————————|
| Gemini 2.5 Pro | 0.967 | 0.788 | 88.76% | $1.25 |
| GPT-5-mini | 0.933 | 0.717 | 78.65% | $0.25 |
| Gemini 2.5 Flash | 0.900 | 0.657 | 78.65% | $0.30 |

Implications for AI Deployment Strategies

Performance Considerations:
Gemini 2.5 Pro remains unrivaled when absolute top performance is required, especially in delicate or highly specialized tasks such as complex SQL querying or in-depth financial analysis.

Cost-Effective High-Volume Usage:
For tasks involving processing millions of tokens—common in enterprise-scale operations—GPT-5-mini offers approximately 94% of Gemini’s performance at