Heidelberg AI News

Video · Bijan Bowen ·

Gemini 4 Hydrogen Is the BEST Flash Model Yet – Full Test!

In short

The video tests Gemini 4 Hydrogen, Google’s new Flash model, across coding, reasoning, game development, robotics, and 3D printing tasks. It performs strongly for its tier, especially in the Subway FPS and Jerry’s apartment 3D-print tests, while using 4% of weekly quota across the test suite.

Key points

  • Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens.
  • The browser OS test was acceptable, with a clean UI, functional apps, and checkpoint snapshots.
  • The C++ skate game was playable but less polished than the best competing models.
  • The robot arm successfully moved a car, but took about one hour at extra-high reasoning.
  • The Subway FPS was the most enjoyable test, featuring a rocket launcher and damageable environments.
  • The Jerry’s apartment model printed accurately without supports on a consumer 3D printer.
  • Agent swarm and extra-high reasoning were used for most tests, consuming 4% of weekly usage.

Written by an AI model from the video’s captions and description.

Watch on YouTube The opening of the article at youtube.com.

In this story

A tool or model opens in the Heidelberg AI Curriculum: what it does, what it costs, how to start. A repo opens on GitHub.

Related stories

All stories

VideoWes Roth

GEMINI 4 is the new era...

The video reviews Google’s Gemini 4 lineup, highlighting Carbon’s multimodal editing and agent capabilities, Hydrogen’s speed and Search integration, Decision-1’s efficient classification, and AlphaProof Code’s formal verification. It argues that Google has re-entered the frontier-model race, while showing that swarm agents can autonomously build and operate complex software systems but may trigger usage safeguards.

VideoWorldofAI

Anthropic SLOWING DOWN!?, Gemini 4 HYDROGEN IS INSANE, OpenAI's NEW Model, & NVIDIA FIRES Jensen!

The video surveys major AI developments, focusing on Gemini 4 Hydrogen’s speed, cost, and agentic coding performance, Anthropic’s safety posture and workplace features, and OpenAI’s unreleased cybersecurity model. It also covers Nvidia leadership and export controversy, new video and computer-use models, and upcoming Microsoft announcements, highlighting rapidly shifting tools and competitive benchmarks for AI builders.