Google DeepMind Introduces Gemini 4 Hydrogen

Google DeepMind introduced Gemini 4 Hydrogen on November 12, describing it as a fast, efficient model for everyday coding and agentic workflows. The company says it is available globally through Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, as well as to paid Google AI Pro and Ultra subscribers in Google Antigravity. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level and 54 at xhigh, compared with 50 for Gemini 4 Argon. Google lists prices of $1.50 per million input tokens and $7.50 per million output tokens across the 1 million-token context window.

Google introduced Orchestrator in Google Antigravity to switch dynamically between models such as Gemini 4 Carbon and Hydrogen during a session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success rates while using 24% less quota on average and finishing 19% faster. In workflows using parallel agents, Carbon can coordinate while Hydrogen serves as worker and verifier agents in isolated workspaces. Compared with Gemini 3.8 Flash, Hydrogen finishes tasks 2.5 times faster at its default high setting and more than triples Terminal-Bench 4.1 scores, from 18.2% to 57%; Google also reports more than 5x fewer fabricated claims. In early partner tests on complex graphs, SQL and data-heavy enterprise work, it ran 2x to 2.8x faster while using 12% to 23% fewer tokens.

Google reports further results across coding, research and business tasks. Hydrogen ranks #5 on Agent Arena, scores 58.8% on HLE-Diamond, ranks #1 on Vals Tax Agent Bench with 52.3%, ties for #1 on the Vals Legal Research Bench with 55.8%, and leads its class on AutomationBench-AA. It scored 53.4% on the Multimodal Knowledge Agent benchmark and 71.6% on Agents’ Last Exam v1.1. Box says its accuracy improved over Gemini 3.8 Flash while running about twice as fast and using 12% fewer tokens. Pixelcut says that across 276 runs Hydrogen completed every design brief, ran 1.8x faster and produced 30% cleaner code.

Hydrogen supports four thinking levels: low, medium, high and xhigh. Google says the model is rolling out globally across developer, enterprise and consumer products. The company reports that it reduces agentic overreach by up to 81% and under-service by up to 79%, cuts misaligned actions and task shortcutting by 12x, and is up to 2x less likely to bypass permission boundaries on impossible or blocked tasks. Google also reports a 96.7% honesty rate in its internal agent honesty evaluation, 99.9% chain-of-thought legibility across more than 200,000 audited turns, and more than 2.3x resilience to indirect prompt injection on the Gray Swan IPI benchmark. False refusals on HealthBench Pro remain under 0.4%, according to Google.

Google DeepMind introduced Gemini 4 Hydrogen on November 12, describing it as a fast, efficient model for everyday coding and agentic workflows. The company says it is available globally through Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, as well as to paid Google AI Pro and Ultra subscribers in Google Antigravity. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level and 54 at xhigh, compared with 50 for Gemini 4 Argon. Google lists prices of $1.50 per million input tokens and $7.50 per million output tokens across the 1 million-token context window.

Google introduced Orchestrator in Google Antigravity to switch dynamically between models such as Gemini 4 Carbon and Hydrogen during a session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success rates while using 24% less quota on average and finishing 19% faster. In workflows using parallel agents, Carbon can coordinate while Hydrogen serves as worker and verifier agents in isolated workspaces. Compared with Gemini 3.8 Flash, Hydrogen finishes tasks 2.5 times faster at its default high setting and more than triples Terminal-Bench 4.1 scores, from 18.2% to 57%; Google also reports more than 5x fewer fabricated claims. In early partner tests on complex graphs, SQL and data-heavy enterprise work, it ran 2x to 2.8x faster while using 12% to 23% fewer tokens.

Google reports further results across coding, research and business tasks. Hydrogen ranks #5 on Agent Arena, scores 58.8% on HLE-Diamond, ranks #1 on Vals Tax Agent Bench with 52.3%, ties for #1 on the Vals Legal Research Bench with 55.8%, and leads its class on AutomationBench-AA. It scored 53.4% on the Multimodal Knowledge Agent benchmark and 71.6% on Agents’ Last Exam v1.1. Box says its accuracy improved over Gemini 3.8 Flash while running about twice as fast and using 12% fewer tokens. Pixelcut says that across 276 runs Hydrogen completed every design brief, ran 1.8x faster and produced 30% cleaner code.

Hydrogen supports four thinking levels: low, medium, high and xhigh. Google says the model is rolling out globally across developer, enterprise and consumer products. The company reports that it reduces agentic overreach by up to 81% and under-service by up to 79%, cuts misaligned actions and task shortcutting by 12x, and is up to 2x less likely to bypass permission boundaries on impossible or blocked tasks. Google also reports a 96.7% honesty rate in its internal agent honesty evaluation, 99.9% chain-of-thought legibility across more than 200,000 audited turns, and more than 2.3x resilience to indirect prompt injection on the Gray Swan IPI benchmark. False refusals on HealthBench Pro remain under 0.4%, according to Google.