Subscribe
AI
2 min read

Artificial Analysis Finds Gemini 4 Neon Third on Its Intelligence Index, Cheaper Than Claude on Coding

Google DeepMind released Gemini 4 Neon, its fourth model in seven weeks, and Artificial Analysis's independent evaluation, published November 18, puts it in third place on the Intelligence Index at 61 (xhigh effort), one point ahead of Gemini 4 Carbon and behind GPT-6.2 Astra (max) and Claude Fable 5.2 (max). The more interesting finding is on price. Intelligence up slightly, cost and latency down Neon (xhigh) costs $6.44 per Intelligence Index task, about 17% less than Carbon's $7.76 at non-

Artificial Analysis Finds Gemini 4 Neon Third on Its Intelligence Index, Cheaper Than Claude on Coding

Google DeepMind released Gemini 4 Neon, its fourth model in seven weeks, and Artificial Analysis's independent evaluation, published November 18, puts it in third place on the Intelligence Index at 61 (xhigh effort), one point ahead of Gemini 4 Carbon and behind GPT-6.2 Astra (max) and Claude Fable 5.2 (max). The more interesting finding is on price.

Gemini 4 Neon cost comparison illustration

Intelligence up slightly, cost and latency down

Neon (xhigh) costs $6.44 per Intelligence Index task, about 17% less than Carbon's $7.76 at non-discounted pricing, and completes a task in roughly 21 minutes against Carbon's 36. The breakdown is worth noting: Neon actually uses about 15% more output tokens per task. The savings come entirely from lower unit prices ($3/$15 per million input/output tokens versus $4/$20) and doubled output speed, around 110 tokens per second against 57.

At high effort, Neon matches Carbon's score at less than half the cost per task, measured against Carbon's standard pricing. That matters for one reason: Google's discounted Carbon price of $2/$10 ends November 30, so the comparison that looks marginal today becomes favorable once list prices apply.

The coding-agent numbers

With Antigravity CLI, Neon (xhigh) scored 59 on the Artificial Analysis Coding Agent Index, three points ahead of Carbon (high, 55) and effectively in line with Claude Code running Claude Fable 5.2 (xhigh, 58) and Claude Opus 5.5 (max, 58). The cost per task splits those near-identical scores apart: $6.86 for Neon, versus $12.61 for Opus 5.5 (max) and $15.47 for Fable 5.2 (xhigh).

On visual and presentation tasks, Neon (high) scored 33.4% on GDP.pdf, five points above Carbon and near the top of measured results, and leads the firm's AA-Briefcase presentation-quality metric at 1771 Elo, 45 points ahead of the previous leader, Claude Sonnet 5.5 (max).

Model details

Neon ships with a 1M-token context window, text, image, video and speech input, four effort levels (low, medium, high, xhigh), and is available through Google's API, the Gemini Enterprise Agent Platform, and to Pro and Ultra subscribers in Antigravity and the Gemini app. Cached input tokens keep the 95% discount ($0.15 per million).

The release cadence is itself a data point. Artificial Analysis frames Neon as the continuation of Google's accelerated shipping schedule, and the lab's position on the leaderboard reflects that strategy: not the highest score, but the best score available at the lowest price on several axes at once. For buyers choosing on price-performance rather than leaderboard position, the November table reads as a genuine squeeze on Anthropic's premium pricing.

Sources

AIGoogleBenchmarks