Development

Gemini 3 Flash's gains traced to agentic RL, at 36% on FrontierMath

development2025-12-20

Claim

Google attributed Gemini 3 Flash's gains to agentic reinforcement learning, scoring 36% on FrontierMath Tiers 1-3 and matching far more expensive models.

Reported metric: 36%

Reporting date: 2025-12-20 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to December 20, 2025 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

About

Reported in

Welcome to December 20, 2025

Actor

Google

Evidence for

The price of reasoning is deflating

Referenced by

Welcome to December 20, 2025

Cite or correct this record

Citation formatCitation text

Gemini 3 Flash's gains traced to agentic RL, at 36% on FrontierMath
Innermost Loop KB — secondary record
Reporting date: 2025-12-20
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2025-12-20-gemini-flash-agentic-rl-frontiermath
Claim: Google attributed Gemini 3 Flash's gains to agentic reinforcement learning, scoring 36% on FrontierMath Tiers 1-3 and matching far more expensive models.
Attributed sources:
- Welcome to December 20, 2025 — https://theinnermostloop.substack.com/p/welcome-to-december-20-2025
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.