Gemini 4 Carbon Claims Top Spot on Intelligence Index v5
Google's latest model leads in reasoning benchmarks while trailing Claude Opus 5.5 in coding.
1Gemini 4 Carbon Performance
Google has launched Gemini 4 Carbon, securing the number one position on the Artificial Analysis Intelligence Index v5. The model achieved a score of 59, demonstrating significant advancements in complex reasoning, scientific tasks, and general knowledge. It currently leads in four out of nine specific index evaluations, including HLE-Diamond and AutomationBench-AA.
2Coding Agent Capabilities
On the Artificial Analysis Coding Agent Index, Gemini 4 Carbon ranks second, trailing only Claude Opus 5.5. With a score of 59, it sits just three points behind the leader. This represents a notable improvement for Google, which has historically underperformed in coding benchmarks, specifically on Terminal-Bench 4.0 and FrontierCode v1.1.
3Agentic Knowledge Improvements
The model shows progress in agentic knowledge work, achieving an AA-Briefcase score of 1609 Elo. This marks a 126-point increase over the Gemini 4 Argon model. Despite these gains, it remains behind competitors like Grok 4.7 and Muse Wonder, with the largest improvements observed in presentation quality and analytical tasks.
4Cost Efficiency Analysis
Gemini 4 Carbon costs $7.76 per Intelligence Index task, which is approximately 23% cheaper than Claude Opus 5.5. While it is more expensive than GPT-6 Astra, it maintains a competitive edge by using 29% fewer output tokens per task. The pricing structure remains consistent with Gemini 4 Argon at $4/$20 per million tokens.
5Omniscience and Hallucination Rates
The model leads the AA-Omniscience v1.1 benchmark with a score of 54, outperforming Claude Opus 5.5 by six points. Accuracy improved by 10 points compared to Argon, reaching 58%. Although the hallucination rate increased from 4% to 11%, it remains the second lowest among all models, significantly better than the 50% rate.
6Availability and Rollout Strategy
Similar to the Gemini 4 Argon release, Carbon is currently limited to selected users and is not yet publicly available. Google is offering a 50% launch discount on standard pricing, though no end date for this promotion has been confirmed. Cached input tokens are also available at a 95% discount rate.
7Index v5 Methodology Updates
The Intelligence Index v5 introduces two new benchmarks and three major upgrades to its evaluation framework. For the first time, the majority of the index is constructed using private datasets. This includes the implementation of Code2Video-AA, which evaluates how effectively models design short product and brand videos from specific design briefs.
8Broader Industry Context
The competitive landscape is shifting as GitHub introduces Copilot Fusion, which orchestrates between Anthropic and OpenAI models to optimize performance and costs. As models like Gemini 4 Carbon continue to narrow the gap in coding and reasoning, the industry focus is increasingly shifting toward balancing high-level intelligence with operational cost efficiency.
Original article: Hacker News — Gemini 4 Carbon Claims Top Spot on Intelligence Index v5