Gemini 4 Neon targets efficiency, but maximum thinking has its own bill

Text
iHOLYSHITVISUAL / TECHAI research breakthroughsiHOLYSHIT / ILLUSTRATION
iHolyShit illustration

Google is rolling out Gemini 4 Neon for complex work. Independent measurements show that its highest effort setting adds one index point while taking substantially longer per benchmark task. Token prices, workload costs and access restrictions tell different parts of the story.

Google began rolling out Gemini 4 Neon on November 18, positioning it around efficient complex work. API pricing is $3 per million input tokens, $15 for output and $0.15 for cached input. But price per token is only part of the bill: thinking effort changes how much work the model does before answering.

The model card says Neon is adapted from Carbon. Text, images, audio and video share a combined million-token input-output window. Default synchronous output is capped at 64K tokens, with a separate background long-decode mode. A large workspace is not an equally large answer allowance. Its knowledge cutoff is March 2026, and some domains may be older.

Artificial Analysis reports an Intelligence Index score of 60 at high effort and 61 at xhigh, with weighted inference-time estimates of 14.8 and 21.5 minutes per index task, calculated from output volume and speed. Xhigh costs a weighted $6.44 per index task. These aggregate measurements are neither a chat-response waiting-time promise nor a flat API fee.

Think of effort as allowing more checks before handing work in. Extra reasoning may help reconcile constraints, but it can consume more tokens and time. The practical bill depends on the path through a job, including intermediate output and retries, not just the unit price. An acceptable wait for overnight analysis may be unsuitable for interactive work.

The general rollout covers the paid API and Pro/Ultra access in Gemini and Antigravity. Eligible Fairwind defenders are due to receive the specialized version over the coming days. The model card notes hallucinations remain possible and says the full safety report is still forthcoming. Teams can choose effort around actual workloads while retaining result checks.

iHolyshit review

The useful choice may be between effort settings, not just model names. Reserving deeper reasoning for the jobs that benefit from it makes more sense than automatically maximizing every setting.

Share this article

What do you think of this story?

Anonymous feedback, up to 5 reactions per article.

Reader comments

Comment without signing in, or sign in to display your member name.

Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.

0/500

No comments yet. Be the first to share.

Sources