Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5

Hacker News by 3 min read 524x views
Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5

Share Post

See example page

Google’s Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5, and second lone to Claude Opus 5.5 on the Artificial Analysis Coding Agent Index

Gemini 4 Carbon scores 59 on Intelligence Index v5, driven by foremost achievement in complex reasoning, knowledge, and specialized tasks. It additionally has ample gains in coding and cognition work, particularly whenever compared to another Gemini models, but it is not foremost in all these areas.

As alongside initiate of Gemini 4 Argon, Gemini 4 Carbon is being rolled out to selected users and is not yet publically available. Google is offering a 50% initiate discount, alongside no confirmed end date. All our charts and figures use norm pricing for the two Gemini 4 Carbon and Argon.

Key results for Gemini 4 Carbon alongside elevated reasoning:

➤ Gemini 4 Carbon leads the Intelligence Index v5, alongside the #1 mark in 4 of the 9 Index evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, AutomationBench-AA. Carbon makes significant gains elsewhere but is not foremost in coding and cognition work

➤ Behind lone Claude Opus 5.5 on coding agents: With Antigravity CLI, Gemini 4 Carbon (high) scores 59 lone 3 points rearward Claude Code alongside Claude Opus 5.5 (max, 62) on the Artificial Analysis Coding Agent Index. Carbon ranks second on the two Terminal-Bench 4.0 (64%) and FrontierCode v1.1 (51%), rearward Claude Opus 5.5 (max) at 56%. Google has historically underperformed in this area and so this betterment stands out.

➤ Improved agentic cognition work, but not yet leading: On AA-Briefcase, Gemini 4 Carbon (high) scores 1609 Elo, a 126-point betterment complete Gemini 4 Argon (high, 1483), but motionless rearward models including Grok 4.7 (xhigh, 1638), Muse Wonder (max, 1659) and Kimi K3.1 Pro (1678). The largest acquire comes from Presentation, up 293 Elo to 1598, alongside a 120-point betterment in Analytical Quality to 1671

➤ Lower Cost per Task than Claude Opus 5.5 for comparable general intelligence: With norm pricing, Gemini 4 Carbon expenses $7.76 per Intelligence Index task, ~23% small than Claude Opus 5.5 (max, $10.04) but ~1.2x GPT-6 Astra (max, $6.25). It uses 123k output tokens per task, 29% small than Claude Opus 5.5 (max, 173k). At the identical $4/$20 norm pricing, its Cost per Task is in row alongside Gemini 4 Argon ($7.68), during scoring 9 points higher on the Intelligence Index

➤ Leading AA-Omniscience v1.1 score, alongside low hallucination: Gemini 4 Carbon scores 54 on our lately updated AA-Omniscience v1.1, 6 points onward of Claude Opus 5.5 (max, 48) and 8 points onward of Gemini 4 Argon (46). Accuracy (% correct) rises 10 points complete Argon to 58%, during its illusion charge rises from 4% to 11%. This is motionless the second lowest of any model, rearward lone Argon, compared alongside 50% for the two Claude Opus 5.5 (max) and GPT-6 Astra (max)

Key example details:

➤ Context Window: 1M tokens

➤ Multimodality: Text, image, video, and address input, alongside content output

➤ Pricing: $4/$20 per 1M input/output tokens, the identical as Gemini 4 Argon’s norm pricing, alongside a 50% discount. Cached input tokens obtain a 95% discount ($0.20 per 1M at norm pricing)

Other Article Hacker News
↑
Close Right Ads
Close Left Ads