AI Research

Google Launches Gemini 4 Hydrogen — Volume-Tier Agentic Model at $1.50/$7.50, 2.5× Faster Than Flash

Google Gemini 4 Hydrogen model launch with three-tier family diagram showing Carbon, Argon, Hydrogen

Google Gemini 4 Hydrogen — Daily Driver for Agentic Workflows at $1.50/$7.50, 2.5× Faster Than Flash

Google launched Gemini 4 Hydrogen on November 12, 2026 — the volume-tier execution model in the Gemini 4 family, positioned beneath Carbon (frontier cyber/long-horizon) and Argon (daily driver). Priced at $1.50/$7.50 per M tokens (input/output), flat across 1M context, Hydrogen delivers 2.5× faster task completion than Gemini 3.8 Flash, 57% on Terminal-Bench 4.1 (vs 18.2% for Flash), and 51 on Intelligence Index (beating Argon's 50 at high, reaching 54 at xhigh).

Bundle context: This is the third Gemini 4 launch in 14 days — Carbon (Oct 28/Nov 10) for cyber/long-horizon, Argon (Sep 30/Oct 13) for daily coding, Hydrogen (Nov 12) for high-volume agentic execution. Together they form a three-tier family mirroring Anthropic's Opus/Sonnet/Haiku.

Model Comparison: The Three-Tier Gemini 4 Family

Tier Model Intelligence Index Terminal-Bench 4.1 Price (Input/Output) Context Role
Frontier Gemini 4 Carbon 59 (#1) ~65% $2/$10 intro → $4/$20 1M Cyber, long-horizon, deep reasoning
Daily Driver Gemini 4 Argon 50 ~57% $2/$10 intro → $4/$20 1M General coding, reasoning
Volume Gemini 4 Hydrogen 51 (high) / 54 (xhigh) 57% $1.50 / $7.50 1M High-volume agentic execution
Prev Gen Gemini 3.8 Flash 31 18.2% $0.75/$3.75 intro 1M Previous volume tier

Key positioning: Hydrogen beats Argon on Intelligence Index (51 vs 50) at default high setting, reaches 54 at xhigh (vs Carbon's 59). Priced at Flash's future standard rate ($1.50/$7.50) — 2× Flash's current intro rate.

Benchmark Performance: Agentic Coding & Reasoning

Benchmark Hydrogen Flash Argon Sonnet 5.5 Notes
Terminal-Bench 4.1 57% 18.2% ~57% 63.9% Agentic coding
Intelligence Index (high) 51 31 50 — Beats Argon
Intelligence Index (xhigh) 54 — — — Near Carbon's 59
HLE-Diamond 58.8% — — — Leads class
Vals Tax Agent Bench 52.3% — — — #1 overall
Vals Legal Research Bench 55.8% — — — Tied #1
Agents' Last Exam v1.1 71.6% — — — Strong agentic
Agent Arena #5 — — — Live benchmark
AutomationBench-AA Leads class — — —
Fabricated claims 5× fewer Baseline — — Reliability
Task speed 2.5× faster Baseline — — Default high setting

Partner validation: Box CEO Aaron Levie reported 74% → 79% accuracy on Complex Work Eval, <½ time, 12% fewer tokens.

Multi-Model Orchestration: Carbon Coordinates, Hydrogen Executes

Orchestrator in Google Antigravity dynamically switches between models in the same session:

  • Carbon = Coordinator (plans, architectural decisions, deep debugging)
  • Hydrogen = Worker/Verifier agents (parallel execution, testing, file edits)
  • Result: Near-Carbon task success, 24% less quota, 19% faster on Terminal-Bench

Analogy: Anthropic's Fable 5.2 as Advisor/Orchestrator + Haiku 5.5 as executor — same pattern, Google's native implementation.

Developer & Enterprise Workflows

Use Case Hydrogen Advantage
Iterative repo coding Surgical edits, parallel file reads, 2–2.8× faster, 12–23% fewer tokens
Debugging Cleaner edits, no unrelated rewrites, recovers from failed tool calls
Open-ended agentic #5 on Agent Arena; leads HLE-Diamond, Vals Tax/Legal benches
Enterprise data/SQL 2–2.8× faster, parallel batching of file reads/searches
Gemini app/Search Better tool calling, less back-and-forth, free-tier auto-selection
Google Sheets Pro/Ultra subscribers; data analysis workflows

Pricing: Flat $1.50/$7.50, Cached at $0.075

Metric Price (per 1M)
Input (uncached) $1.50
Output $7.50
Implicit cache read $0.075 (5% of input)
Explicit cache Not supported
Batch Not published
Context window 1M tokens flat
Max output 65,536 tokens (sync)

vs. Flash: 2× Flash's current intro ($0.75/$3.75), matches Flash's scheduled standard rate from Jan 1, 2027. vs. Carbon/Argon: 25–60% cheaper; Carbon/Argon intro $2/$10 until Nov 30, then $4/$20. vs. Competition: Haiku 5.5 (~$0.25/$1.25), Sonnet 5.5 ($4/$20), Luna (~$0.15/$0.75) — Hydrogen mid-tier pricing, frontier-class agent scores.

Availability: Global Rollout Across All Surfaces

Platform Access
Gemini API / AI Studio Paid tier; Pro/Ultra in Antigravity
Android Studio Paid tier
Google Antigravity Pro/Ultra subscribers
Gemini for Business API/model dropdown, Antigravity, CodeMender
Gemini App Model dropdown; free users via auto-selection
Google Search All users globally
Gemini in Sheets Pro/Ultra subscribers
Free tier Auto model selection in Gemini app

No explicit cache, no batch pricing published yet.

Competitive Landscape: The Three-Way Volume Tier Race

Model Provider Price (In/Out) Index Terminal-Bench Context Status
Hydrogen Google $1.50/$7.50 51/54 57% 1M Live Nov 12
Haiku 5.5 Anthropic ~$0.25/$1.25 ~45 ~50% 200K Live
Sonnet 5.5 Anthropic $4/$20 58 ~64% 200K Live
Luna OpenAI ~$0.15/$0.75 ~45 ~45% 128K Live
Muse Ember Meta $0.10/$0.50 36 — 1M Live Nov 9
DeepSeek V4.1 Flash DeepSeek ~$0.10/$0.40 ~34 — 128K Live (open)
GLM 5.3 Flash Zhipu Competitive — — 128K Live

Hydrogen's niche: Frontier-class agent scores (51-54 Index, 57% Terminal-Bench) at mid-tier pricing — between Haiku/Ember (cheaper, lower scores) and Sonnet/Carbon (pricier, higher scores).

Strategic Significance: Three-Tier Family Complete

Launch Date Model Purpose
1. Carbon Oct 28 / Nov 10 Frontier cyber/long-horizon #1 Index (59)
2. Argon Sep 30 / Oct 13 Daily driver coding Solid baseline (50)
3. Hydrogen Nov 12 High-volume execution Speed/efficiency (51 Index, 57% TB)

Google's three-tier velocity: 3 models in 43 days (Sep 30 → Nov 12) — matches Anthropic's Opus/Sonnet/Haiku cadence but compressed. Orchestrator enables native multi-model workflows.

What Comes Next

  1. Batch pricing — Critical for high-volume offline workloads
  2. Explicit cache support — Currently only implicit
  3. Higher sync output cap — 65K tokens limits long generations
  4. Independent benchmarks — Artificial Analysis, third-party validation
  5. OpenAI/Anthropic response — Expected Luna/Haiku price cuts or speed improvements
  6. Orchestrator adoption — Multi-model workflow uptake in Antigravity
  7. Enterprise migration — Flash → Hydrogen transition for 3.8 Flash users

The Bigger Picture

Gemini 4 Hydrogen completes Google's three-tier agentic family — Carbon plans, Argon reasons, Hydrogen executes at scale. The Orchestrator makes multi-model workflows native, not manual.

Pricing at Flash's future standard ($1.50/$7.50) signals Google's cost curve confidence — Hydrogen delivers frontier-class agent scores at volume pricing. For teams running thousands of agent tasks daily, the 2.5× speed + 57% Terminal-Bench + mid-tier price is a compelling value proposition.

If Orchestrator adoption takes off, Google owns the full agentic stack: planner (Carbon), reasoner (Argon), executor (Hydrogen), coordinator (Orchestrator) — a vertically integrated agentic platform no rival yet matches.


Related AIPress Coverage

Sources

  • Google Blog (Nov 12, 2026): "Introducing Gemini 4 Hydrogen for Agentic Workflows" (Koray Kavukcuoglu)
  • 9to5Google (Nov 12, 2026): "Google announces Gemini 4 Hydrogen with 'significant leaps'"
  • MarkTechPost (Nov 13, 2026): Detailed benchmark analysis
  • VentureBeat (Nov 12, 2026): "Google launches Gemini 4 Hydrogen — a faster, smarter, pricier Flash successor"
  • Business Standard (Nov 13, 2026): Features, pricing, availability
  • AI Daily (Nov 12, 2026): Three-step rollout timeline
  • Indie Hackers (Nov 2026): Hydrogen vs Argon Intelligence Index comparison
  • AI TLDR (Nov 2026): Model card summary
  • Box CEO Aaron Levie (Nov 2026): Partner validation on X
  • Artificial Analysis Intelligence Index v5.0.1 (Nov 2026)

Building something with AI?

DevsIsle designs and ships AI systems, agents and integrations for teams that need it done properly.

Talk to our team →