One spot above Opus 5.5 at half the cost per task: Gemini 4 Neon's first day on outside leaderboards
First seen on X 15 hours ago@arena ♥ 46022 views
Google's Gemini 4 Neon, released Nov. 18, climbed outside leaderboards within hours. On Arena's Agent Arena Index, which ranks models on 4,200 real user agent sessions, it placed second with 1,336 points. Its median cost per task of $2.82 was 53% lower than third-place Claude Opus 5.5 and 69% lower than first-place Claude Fable 5.2.
On the Vals Index from evaluation firm Vals AI, Neon scored 70.9%, 0.4 points behind Fable 5.2 for second place, at less than half the cost per task, and took first on its Finance and Legal indexes and Tax Agent. Google says it scores 61 on the Artificial Analysis Intelligence Index, third overall, and ranks first on the Coding Agent Index. API pricing is $3 per million input tokens and $15 per million output tokens. Google AI Pro and Ultra subscribers can pick it in the Gemini app, and GitHub added it to Copilot Pro+, Max, Business and Enterprise the same day.
Hands-on results followed. Developer notjazii ran the same prompt at the highest reasoning setting and reported about $3.50 and 20 minutes for Neon versus about $7.20 and 18 minutes for Opus 5.5. Others were skeptical, asking whether winning every reported benchmark really makes it better, and one user said a first test "does NOT look promising." Some pointed out that Gemini 4 Carbon stays on a $2/$10 promotional price through Nov. 30, so Carbon is cheaper until then.
Neon is Google's fourth new Gemini 4 model in a month (The Decoder). Google places it between the top Carbon model and the lighter Hydrogen, saying it responds faster than Carbon and makes fewer unrequested code changes.
Comments 0
Select a sentence in the story (long-press on phones) to comment on it.
“Big news: Gemini 4 Neon (High) is the best @GoogleDeepmind model on the Agent Arena Index, ranking #2 overall and reshaping the Pareto frontier!”
