The AI race has seen another core model update. Google officially released its new large model, Gemini4Hydrogen, on Thursday local time. As a new-generation flagship model for both developers and everyday users, it has achieved significant leaps in running speed, coding capabilities, and agent task handling.
Major Performance Leap: 2.5x Faster, Coding Benchmarks Jump Sharply
In terms of overall performance, compared with the previous Gemini3.8Flash, Gemini4Hydrogen shows all-around evolution:
Running speed: Under the default high-reasoning mode, the new model completes tasks 2.5 times as fast as the previous generation, allowing it to handle complex workflows with greater efficiency.
Coding ability: In the rigorous Terminal-Bench4.1 test, its score surged from 18.2% previously to 57%, marking a qualitative leap in coding and terminal operation capabilities.
Lower hallucination rate: The model produces significantly fewer fabricated statements (hallucinations) in real-world operation, making its outputs more solid and reliable.
Inclusive Access and Pricing Strategy: Free Tier Automatically Covered, API Pricing Announced
In terms of user reach and ecosystem openness, Google adopted a deployment strategy that serves both the general public and professional developers:
Regular user experience: Free users of the Gemini app will, through the automatic model selection feature, automatically invoke the Hydrogen model in prompt scenarios the system deems appropriate.
Long text and API pricing: The model supports up to 1 million tokens of input and up to 1 million tokens of output; for API call pricing, input costs $1.50 per million tokens, and output costs $7.50 per million tokens.
At the same time, Google also launched a new Orchestrator feature, which supports dynamic task allocation and coordination across different models, further optimizing the efficiency of handling complex, long multi-step tasks.
