SAN FRANCISCO — Google DeepMind’s latest frontier AI model, Gemini 4 Carbon, has claimed the top spot on the Artificial Analysis Intelligence Index v5 with an aggregate score of 59.
The benchmark result places Google’s newest release one point ahead of Anthropic’s Claude Opus 5.5, which holds a score of 58, and two points ahead of OpenAI’s GPT-6 Astra at 57.
However, while Gemini 4 Carbon leads the overall benchmark index, initial access remains strictly controlled, and competitor models continue to hold leads across specific domain tasks.
Intelligence Index Standings and Performance Breakdown
The Artificial Analysis Intelligence Index v5 evaluates large language models across multiple foundational benchmarks, including reasoning, science, and plain-API coding evaluations.
Artificial Analysis Intelligence Index v5 Standings
+-------------------------------------------------------------------------+
| MODEL | DEVELOPER | INDEX SCORE |
+-------------------------------------------------------------------------+
| Gemini 4 Carbon | Google DeepMind | 59 |
| Claude Opus 5.5 | Anthropic | 58 |
| GPT-6 Astra | OpenAI | 57 |
| Gemini 4 Argon | Google DeepMind | 50 |
+-------------------------------------------------------------------------+
Gemini 4 Carbon showed notable gains over its predecessor, Gemini 4 Argon, advancing nine points on the aggregate index. The model demonstrated strong performance in long-context processing, scientific problem-solving, and automated cybersecurity evaluations.
Task-Specific Divergence Across Competitors
Despite Carbon’s lead on the overall index, evaluation data shows that competing frontier models retain advantages in specific operational domains:
Task-Specific Benchmark Comparisons
+-------------------------------------------------------------------------+
| CODING AGENTS (Coding Agent Index) |
| * Claude Opus 5.5 leads with 62 points compared to Carbon's 59 points. |
+-------------------------------------------------------------------------+
| COMPUTER CONTROL & OS INTEGRATION (OSWorld) |
| * Claude Opus 5.5 maintains top performance in direct desktop and OS |
| navigation workflows. |
+-------------------------------------------------------------------------+
| PROJECT-SCALE SOFTWARE ENGINEERING (FrontierSWE) |
| * OpenAI's GPT-6 Astra holds the leading position for large-scale |
| repository software engineering tasks. |
+-------------------------------------------------------------------------+
While Carbon outperformed Opus 5.5 in plain-API terminal benchmark runs, specialized coding setups using Anthropic’s agentic framework kept Opus 5.5 ahead in practical developer tools.
Restricted Access and Deployment Strategy
Unlike broadly accessible commercial releases, deployment for Gemini 4 Carbon is initially restricted.
Due to the model’s advanced autonomous capabilities and focus on cybersecurity workflows, Google DeepMind has limited early access to select cyber defense partners and specialized research organizations undergoing pre-deployment testing protocols. Standard commercial users and developers continue to rely on Gemini 4 Argon and competing commercial APIs while Google prepares Carbon for wider release.
