Gemini 4 Hydrogen
- Framework: Frontier Safety Framework (April 2026)
- 6 statements about capability levels
- 7 tests by the provider
- 2 source documents
What the provider states
- Autonomy & replication · CCL (Critical Capability Level) · Stated as not crossed
CCL not reached (carried forward from Argon) .
Google DeepMind · 12 November 2026 · Read the source - Chemical, biological, radiological and nuclear · CCL (Critical Capability Level) · Stated as not crossed
CCL not reached, following additional targeted testing.
Google DeepMind · 12 November 2026 · Read the source - Chemical, biological, radiological and nuclear · TCL (Tracked Capability Level) · Stated as not crossed
TCL not reached (carried forward from Argon).
Google DeepMind · 12 November 2026 · Read the source - Cyber · CCL (Critical Capability Level) · Stated as not crossed
CCL not reached, following additional targeted testing.
Google DeepMind · 12 November 2026 · Read the source - Deception & sabotage · CCL (Critical Capability Level) · Stated as not crossed
CCL not reached (carried forward from Argon).
Google DeepMind · 12 November 2026 · Read the source - Harmful content · CCL (Critical Capability Level) · Stated as not crossed
CCL not reached (carried forward from Argon).
Google DeepMind · 12 November 2026 · Read the source
The provider's own tests
- Agentic overreach, under-service, and misaligned action evaluations across tool-use and coding · The provider · Deception & sabotage
Reported result: Overreach reduced by up to 81%; under-service reduced by up to 79%; misaligned actions and task shortcutting reduced 12x
It reduces overreach in agentic tasks by up to 81% and under-service by up to 79%, and cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12x.
Google DeepMind · 12 November 2026 · Read the source - Chain-of-thought legibility audit · The provider · Deception & sabotage
Building on our commitment to reasoning transparency, Hydrogen also preserves 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns.
Google DeepMind · 12 November 2026 · Read the source - Gray Swan IPI benchmark · The provider · Injection & agentic
Against indirect prompt injection attacks, Hydrogen is more than 2.3x more resilient on the Gray Swan IPI benchmark, while significantly improving robustness across tool use, computer use, and coding agents.
Google DeepMind · 12 November 2026 · Read the source - Image to Text Safety · The provider · Harmful content
Reported result: +7.00pp versus Gemini 3.8 Flash
Image to Text Safety (higher is better) Automated content safety evaluation measuring adherence to safety policies with both images and text as inputs +7.00pp
Google DeepMind · 12 November 2026 · Read the source - Internal agent honesty evaluation and permission-boundary test on blocked tasks · The provider · Deception & sabotage
When faced with impossible or blocked tasks, it is up to 2x less likely to bypass permission boundaries and achieves a 96.7% honesty rate on our internal agent honesty evaluation.
Google DeepMind · 12 November 2026 · Read the source - Multilingual Safety · The provider · Harmful content
Reported result: +1.10pp versus Gemini 3.8 Flash
Multilingual Safety (higher is better) Automated safety policy evaluation across multiple languages +1.10pp
Google DeepMind · 12 November 2026 · Read the source - Text to Text Safety · The provider · Harmful content
Reported result: +0.80pp versus Gemini 3.8 Flash
Text to Text Safety (higher is better) Automated content safety evaluation measuring adherence to safety policies across text prompts +0.80pp
Google DeepMind · 12 November 2026 · Read the source
No document we read states: a release date, an independent or government test, a safeguard. No incident on record names this version.
Documents listed for this version
- Gemini 4 Hydrogen — Model Card · Google DeepMind · 12 November 2026
- Gemini 4 Hydrogen · Google DeepMind · 12 November 2026