AI Trends

Google Releases Gemini 4 Neon, Ranking Third on a Third-Party Intelligence Index

Sina TechnologyUpdated Nov 19 09:41

AI summary

Google has released Gemini 4 Neon, which ranks third on the Artificial Analysis Intelligence Index with a score of 61, surpassing Gemini 4 Carbon's score of 60. The model's API is priced at $3 per million input tokens and $15 per million output tokens. It is available to Google AI Pro and Ultra members, while the free API tier is not currently offered.

Why read this

SelectedAI score 82

The material compiles multiple third-party evaluations, API pricing details, and membership availability, and can be used to compare the performance and cost of Neon with peer models.

Topics

Tags

Body · AI translation

Add new public beta promotional copy Add new public beta promotional copy end Add app-exclusive traffic redirection

(Source: AI Information Gap)

Google has just released a new model, Gemini 4 Neon. This mid-sized model scored 61 on the Intelligence Index from third-party evaluator Artificial Analysis, ranking third overall and beating Google’s own flagship Gemini 4 Carbon by one point.

This time, subscribers to the $19.99-per-month Google AI Pro plan also get access, with a gradual rollout in the Gemini App starting today.

Google has recently entered a diarrhea-like mode of nonstop updates.

On November 11, Gemini 4 Carbon became generally available. The next day, Google released its first decision model, Gemini Decision 1. Another day later, Gemini 4 Hydrogen arrived as well.

Today, the world model Genie 4 had barely opened to Google AI Ultra subscribers when, just a few hours later, Gemini 4 Neon was officially launched.

Come on—time to learn some more English words.

The original hierarchy of Google’s Gemini 4 family looked like this.

Carbon is the large model, reserved for the most difficult long-running tasks and deep research; Hydrogen is the small model, responsible for everyday coding; and Neon is the mid-sized model between the two. The name is easy enough to remember: Neon, pronounced “nee-on,” is the gas used in neon lights.

With Gemini 4 Neon running at the highest reasoning intensity, xhigh, it scored 61 on Artificial Analysis’s Intelligence Index, ranking third overall behind only Claude Fable 5.2 and GPT-6.2 Astra.

Carbon scored 60, Claude Opus 5.5 scored 59, and GPT-6.2 Sol scored 57.

On the new agent leaderboard from the anonymous large-model evaluation platform Arena, the mid-sized Neon also surpassed the large Carbon.

Gemini 4 Neon scored 1336 and ranked second, while Carbon placed only fourth.

Compared with third-place Claude Opus 5.5, Neon scored one point higher while costing 53% less per task.

On evaluator Vals AI’s leaderboard, Neon also debuted in second place with a score of 70.9%, just 0.4 percentage points behind Claude Fable 5.2.

The key point: its cost per task is less than half that of Fable.

Neon ranks first on both the finance and legal sub-leaderboards.

Gemini 4 Neon’s API pricing is $3 per million input tokens and $15 per million output tokens.

Carbon’s regular pricing is the same as Claude Opus 5.5: $4 per million input tokens and $20 per million output tokens. Neon is 25% cheaper than both.

However, Carbon currently has introductory pricing of $2 per million input tokens and $10 per million output tokens, which does not end until November 30.

The Google AI Pro subscription may offer the best value of all.

When Carbon opened up last week, Pro subscribers were excluded. This time, both Pro and Ultra subscribers can select Neon from the model menu in the Gemini App. The Tasks feature in Gemini, which handles miscellaneous multistep work for you, can now also be assigned to Neon. Google says it works faster than larger models and lets the same usage allowance last longer.

PS: The Tasks feature currently found in the Gemini App is the former Gemini Spark.

Antigravity is available only to paying Pro and Ultra subscribers. Developers who want to access Neon through the API must likewise use a paid tier; Neon is not available on the free tier.

In the official scorecard released by Google, Gemini 4 Neon was compared with Claude Opus 5.5 and GPT-6.2 Sol across 19 tests and took first place in 16 of them.

Of the three tests Neon lost, two were programming tests where it fell behind Claude Opus 5.5 by about one percentage point in each, while the other was a chart-reading test that it lost to GPT-6.2 Sol.

The gap was even larger on GraphWalks, an ultra-long-context test spanning 256,000 to one million tokens.

Neon scored 95.2%, while Claude Opus 5.5 managed only 66.8%.

A new model release would not be complete without a knockout chart.

The chart below comes from Google. The horizontal axis shows cost per task, while the vertical axis represents model intelligence—the farther toward the upper left a model sits, the more formidable it is.

At the default high setting, Google says Neon costs 32% less per task than Carbon.

Thank you for reading this far. I’m Mu Yi, helping you see what’s coming in AI one step ahead every day. If you found this article useful, please give it a like, tap “Watching,” and share it with friends who use AI. To receive updates as soon as they arrive, add me to your starred accounts ⭐. See you tomorrow. ]article_adlist Statement: This is an exclusive Sina.com article and may not be reproduced without authorization. tech_web_article_v2015_block_bottom Non-targeted 300*250 button 17/09 wenjing begin

Non-targeted 300*250 button end

Source

Sina Technology