Fastest AI Models -- Speed Rankings
When response time matters. Real-time applications, chatbots, and interactive tools need speed above all. Ranked by tokens per second output speed.
Our Top Pick
Mercury 2 delivers the fastest output at 894 tok/s, ideal for real-time and interactive applications.
Methodology
Ranked by output speed in tokens per second, measured across standardized prompts across standardized evaluations.
| # | Model | Provider | Speed | Latency | Intelligence | Price/1M | |
|---|---|---|---|---|---|---|---|
| 1 | I | Mercury 2 | Inception | 894 tok/s | 3.81s | 32.8 | $0.25 |
| 2 | I | Granite 3.3 8B (Non-reasoning) | IBM | 402 tok/s | 9.53s | 7.0 | $0.03 |
| 3 |  | Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning) | 389 tok/s |
Frequently Asked Questions
What is the best AI for real-time applications in 2026?
Based on our benchmark rankings, Mercury 2 is currently the top-ranked model for real-time applications. See the full rankings above for alternatives.
How much does Mercury 2 cost for real-time applications?
Mercury 2 costs $0.25/1M input tokens. For 100 requests per day at 2,000 tokens each, that's approximately $1.50/month.
Is there a free AI model good for real-time applications?
Currently, there are no free models that rank highly for real-time applications. Check our free models page for the best zero-cost options.
Not sure which model to pick?
Take our 30-second quiz and get a personalized recommendation.
Take the Quiz