The Extended Brief

Google rolls out Gemini 4 Hydrogen, finally replacing 3.8 Flash

Brief by The AI News AI newsroom · Nov 12, 2026, 3:11 PM EST edition

Original reporting by Ars Technica AI — Ryan Whitwam · published Nov 12, 2026, 2:23 PM EST

Google's default fast model tier just changed hands: Hydrogen replaces 3.8 Flash across all Google products, with claimed jumps in speed, coding, and factual reliability.

Key points

• Google has replaced Gemini 3.8 Flash with Gemini 4 Hydrogen across its products, with access beginning immediately. source ↗
• Google claims Hydrogen runs 2.5 times faster than Flash 3.8 even in high thinking mode. source ↗
• Google's analysis says Hydrogen reduces fabricated claims by more than 5x versus the previous Flash model. source ↗
• Google says Hydrogen more than tripled 3.8 Flash's score on the Terminal-Bench code-generation benchmark. source ↗
• Hydrogen scored 54 on the Artificial Analysis Intelligence Index in xhigh mode, ahead of Argon's 50. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Benchmark Hydrogen against your existing Flash 3.8 workloads now, since the replacement is already live across Google's products.
• Test the new xhigh thinking level on your hardest coding tasks, given Google's claimed tripling of Terminal-Bench scores.
• If you pay for Google's most expensive subscription mainly for early model access, reassess that tier now that Hydrogen launched broadly instead of to top-tier subscribers first.
• Treat the 5x fabricated-claims reduction as a vendor figure and measure factual accuracy on your own data before relying on it.

Context

Gemini is Google's flagship AI model family, previously organized into tiers including the fast, economical Flash line that most users in its ecosystem encounter. Recent releases use element names like Argon, Carbon, and Hydrogen, and offer adjustable thinking levels that trade speed for deeper reasoning. Terminal-Bench and the Artificial Analysis Intelligence Index are external benchmarks measuring coding ability and overall model quality.

What to watch

• Independent replication of Google's speed and fabrication claims by third-party benchmarkers.
• Whether Hydrogen's xhigh mode keeps edging out Argon on benchmarks beyond the Artificial Analysis Intelligence Index.

Related briefs

• Praxis-1 - a runwayml Collection
• Anthropic tiptoes into MS Office territory
• DeepMind reveals Gemini Robotics 2.5 and open safety benchmark "ASIMOV-jailbreak"
• Claude Fable 5.2 Scores 65 on Artificial Analysis, Highest So Far

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.