Unverified50% confidenceFactExact time
In Gemini 4 Neon's output, 56% to 81% of the budget is used for deep thinking (the reasoning process).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/20/2026
First Seen
Valid until: 2/18/2027
Sources
Gemini 4 Neon Hands-On: Value King or Confident Catastrophe?
bilibili闲人与杂谈11/19/2026
Related Entities
Related Claims
UnverifiedIn the Anti-Gravity CLI test, Gemini 4 Neon spent 56% to 81% of its tokens on thinking, and with no opening filler it gave the answer directly.79% similarUnverifiedGemini 4 Neon's retrieval accuracy on million-level ultra-long contexts is 95.2%.78% similarPartially VerifiedGemini 4 Neon scored 61 on the Artificial Analysis Intelligence Index at the xhigh thinking level, ranking third overall.76% similarUnverifiedLowering Gemini 4 Neon's thinking depth from very high to the upper-middle setting reduced its overall score from 61 to 60, cut output tokens by 36.6%, and lowered the average cost per comprehensive evaluation task from $6.44 to $3.74.76% similarUnverifiedGemini 4 Neon outputs at most 65,536 tokens per request in the default synchronous mode, and enabling the long-output feature expands this to as large as its total capacity.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1190405API
curl https://kongchang.com/api/v1/knowledge/claims/1190405MCP
get_claim(id=1190405)