Gemini 4 Carbon Intelligence Index v5 Release Notes

Gemini 4 Carbon Performance Overview

Google’s Gemini 4 Carbon has secured the #1 position on the Artificial Analysis Intelligence Index v5, achieving a score of 59. The model demonstrates significant gains in complex reasoning, scientific tasks, and knowledge-based work compared to previous Gemini iterations. While it leads in four of the nine index evaluations—including HLE-Diamond and Terminal-Bench-Science—it currently ranks second to Claude Opus 5.5 on the Artificial Analysis Coding Agent Index.

Technical Capabilities and Benchmarks

Gemini 4 Carbon shows marked improvements in agentic knowledge work, specifically within the AA-Briefcase evaluation, where it achieved a 126-point Elo increase over the Gemini 4 Argon model. Despite these gains, it remains behind models such as Grok 4.7 and Muse Wonder in certain knowledge-intensive tasks.

In coding benchmarks, Carbon scored 59 on the Artificial Analysis Coding Agent Index, trailing Claude Opus 5.5 (max), which scored 62. However, this represents a notable improvement for Google, as the company has historically underperformed in coding-specific agentic tasks. The model features a 1M token context window and supports multimodal inputs including text, image, video, and speech.

Cost and Availability

Gemini 4 Carbon is currently undergoing a phased rollout to selected users and is not yet publicly available. Google has implemented a 50% launch discount on the model's standard pricing of $4 per 1M input tokens and $20 per 1M output tokens. When compared to Claude Opus 5.5, Carbon offers a lower cost per task, estimated at $7.76 compared to $10.04, while maintaining comparable intelligence metrics.

Community Reception and Concerns

The announcement has been met with skepticism from the developer community, primarily due to the model's limited availability. Many users expressed frustration that the model cannot be independently tested, with some noting that previous Google models have suffered from performance degradation or "lobotomization" shortly after release.

Discussion on the Artificial Analysis index methodology also emerged, with some observers questioning the timing of index updates relative to major model releases. Critics raised concerns regarding the transparency of the benchmarks, noting that a significant portion of the evaluations are private and cannot be replicated by the public.

"There's an issue with this index where many models are listed in the model chooser, but missing from the index. This makes it a bit tricky to see the improvements over time." — @terhechte

"The benchmarks of a model that hasn't been released yet are definitely suspicious. I also think Google should stop 'releasing' models that aren't actually available." — @tlogan

"As of their latest update, I find their index hard to trust, and a bit shady. 60% of the benchmarks contributing to the AA index are now private, nobody can replicate them." — @GodelNumbering