Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5

Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5

Google’s Gemini 4 Carbon ranks first on the Artificial Analysis Intelligence Index v5, while placing second behind Claude Opus 5.5 on the Artificial Analysis Coding Agent Index.

Gemini 4 Carbon scores 59 on the Intelligence Index v5. Its result is supported by leading performance in complex reasoning, knowledge, and scientific tasks. The model also makes significant gains in coding and knowledge work compared with other Gemini models, although it does not lead every category.

Like Gemini 4 Argon, Gemini 4 Carbon is being released to selected users and is not yet publicly available. Google is offering a 50% launch discount, with no confirmed end date. The figures below use standard pricing for both Gemini 4 Carbon and Gemini 4 Argon.

Main results

First place in four Intelligence Index evaluations

Gemini 4 Carbon leads the Intelligence Index v5 and records the highest score in four of its nine evaluations:

• HLE-Diamond
• Terminal-Bench-Science
• AA-Omniscience
• AutomationBench-AA

The model improves substantially in other areas, but it does not lead the coding and knowledge-work evaluations.

Second place on coding-agent benchmarks

With Antigravity CLI, Gemini 4 Carbon with high reasoning scores 59 on the Artificial Analysis Coding Agent Index. That is three points behind Claude Code with Claude Opus 5.5 at maximum reasoning, which scores 62.

Gemini 4 Carbon ranks second on both of the underlying benchmarks:

• Terminal-Bench 4.0: 64%
• FrontierCode v1.1: 51%

Claude Opus 5.5 at maximum reasoning scores 56% on FrontierCode v1.1. The improvement is notable because Google’s models have historically performed less strongly in this area.

Gains in agentic knowledge work

Gemini 4 Carbon with high reasoning scores 1609 Elo on AA-Briefcase. This is a 126-point improvement over Gemini 4 Argon with high reasoning, which scores 1483.

Despite the improvement, Carbon remains behind several models, including:

• Grok 4.7 with extra-high reasoning: 1638
• Muse Wonder at maximum reasoning: 1659
• Kimi K3.1 Pro: 1678

The largest improvement is in the Presentation category, where Carbon gains 293 Elo points to reach 1598. Analytical Quality improves by 120 points to 1671.

Cost per task

At standard pricing, Gemini 4 Carbon costs $7.76 per Intelligence Index task. This is approximately 23% less than Claude Opus 5.5 at maximum reasoning, which costs $10.04, but about 1.2 times the $6.25 cost of GPT-6 Astra at maximum reasoning.

Gemini 4 Carbon uses 123,000 output tokens per task, 29% fewer than Claude Opus 5.5 at maximum reasoning, which uses 173,000 tokens.

At the same standard pricing of $4 per 1 million input tokens and $20 per 1 million output tokens, Carbon’s cost per task is close to Gemini 4 Argon’s $7.68. Carbon scores nine points higher on the Intelligence Index.

AA-Omniscience performance and hallucination rate

Gemini 4 Carbon scores 54 on AA-Omniscience v1.1, six points ahead of Claude Opus 5.5 at maximum reasoning, which scores 48, and eight points ahead of Gemini 4 Argon, which scores 46.

Carbon’s accuracy, measured as the percentage of correct answers, increases by 10 percentage points over Argon to 58%. Its hallucination rate also rises, from 4% to 11%.

At 11%, Carbon has the second-lowest hallucination rate among the models listed, behind Gemini 4 Argon. Claude Opus 5.5 at maximum reasoning and GPT-6 Astra at maximum reasoning each have a 50% hallucination rate.

Model specifications

• Context window: 1 million tokens
• Input modalities: Text, image, video, and speech
• Output modality: Text
• Standard pricing: $4 per 1 million input tokens and $20 per 1 million output tokens
• Cached input pricing: $0.20 per 1 million tokens at standard pricing, representing a 95% discount
• Launch pricing: 50% discount