Overview
Gemini 4 Hydrogen is the third model in Google's Gemini 4 line, released on November 12, 2026 under the API name gemini-4-hydrogen. It does not replace Carbon or Argon. Those two handle long-horizon reasoning; Hydrogen is built to finish a task quickly and act as the execution end of an agent loop. Google positions it as the upgrade path for teams still running Gemini 3.8 Flash.
Speed is the headline. At the default high setting it finishes tasks 2.5 times faster than Gemini 3.8 Flash, and Terminal-Bench 4.1 rises from 18.2% to 57%, which Google describes as more than tripling. Fabricated claims drop by more than fivefold, which matters more for work where the model has to decide whether it should commit to an answer at all. On the Artificial Analysis Intelligence Index it scores 51 at the default level and 54 at xhigh, against 50 for its sibling Gemini 4 Argon.
Context and pricing are the other draw. It takes native 1 million-token input and returns up to 1 million output tokens, priced flat across the entire window: $1.50 per million input tokens and $7.50 per million output tokens. The usual step-up pricing for long context does not appear in this generation.
When to switch back to the other two. Argon and Carbon sit at 2 dollars input and 10 dollars output during the introductory period, doubling to 4 and 20 on December 1, 2026, while Hydrogen stays at 1.50 and 7.50 across the full window, so the same workload costs roughly a quarter to a third as much here. What that saves in money it gives up in reasoning depth, and work that needs extended argument or runs across days belongs back on Carbon or Argon. The stretch that reads context and acts on it, which is the bulk of any agent loop and the last place worth paying a reasoning premium, is exactly where Hydrogen fits.
Key Features
- Four thinking levels: low, medium and high carry over with high as the default, and xhigh is new in this generation, reserved for the hardest problems. The pitch is one model covering everything from a fast reply to deep reasoning, picked by level rather than by switching models.
- Million-token input and output: Native 1 million-token input, with output capped at 1 million tokens as well. That output ceiling is what makes long reports, large code drops and entire contracts writable in one pass.
- Terminal-Bench 4.1 from 18.2% to 57%: The largest single jump in this generation. The benchmark tests whether a model can run a sequence of shell commands correctly, which is the closest public proxy for agent work.
- Fivefold fewer fabricated claims: Google reports more than a 5x reduction in fabricated claims. For tasks where the model must judge whether a conclusion is warranted, this counts for more than a raw score.
- Flat pricing across the window: One rate for the whole 1 million-token context: $1.50 per million input tokens and $7.50 per million output tokens, with no length tiers. Long-context cost becomes a number you can compute in advance.
- Paired with Carbon in Antigravity: Orchestrator switches between Carbon and Hydrogen inside a single session instead of using one model end to end. Google reports near-Carbon task success on Terminal-Bench 4.0 while consuming 24% less quota on average and finishing 19% faster. Carbon can act as coordinator while Hydrogen drives worker and verifier agents building and testing in isolated workspaces.
- Benchmark results: HLE-Diamond 58.8%, Vals Tax Agent Bench 52.3% for first overall, Vals Legal Research Bench 55.8% tied for first, Agents' Last Exam v1.1 71.6%, Multimodal Knowledge Agent 53.4%, and fifth on the live Agent Arena leaderboard.
- Every surface on day one: The Gemini app model dropdown, Google Search, the Gemini API and AI Studio, Gemini for business, Google Antigravity and Gemini in Sheets all opened on the same day. Free users reach it through auto model selection in the app rather than picking it directly.
Use Cases
- Serving as the execution model in an agent loop while Carbon or Argon handles planning
- Reading an entire codebase or several hundred pages in one pass before answering
- High-volume calls where latency matters, using the low or medium level to hold cost down
- Search and everyday Q&A, where free users are routed to it automatically in the app
- Spreadsheet and office work through Gemini in Sheets
Pros
- Four thinking levels cover instant replies through deep reasoning without swapping models
- Flat pricing on 1M input and 1M output makes long-context budgeting predictable
- Works alongside Carbon in the same session while spending less quota
- App, Search and API all opened the same day, so migration needs no waiting
- The Terminal-Bench gain shows up directly in agent tasks
Pricing
Priced flat across the full 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens, with no tiers. In the Gemini app, free users reach it through auto model selection, while Google AI Pro and Ultra subscribers can pick it from the model dropdown. Gemini in Sheets starts with Pro and Ultra. Developers use the Gemini API and AI Studio, and Gemini for business plus Google Antigravity are available the same day.
Summary
Gemini 4 Hydrogen completes the execution side of Gemini 4: 1 million tokens in and out, flat pricing across the window, 2.5x faster than Gemini 3.8 Flash at the default level, and Terminal-Bench 4.1 at 57%. Its division of labor with Carbon and Argon is shipped as a product feature through Antigravity's Orchestrator, which switches models mid-session and, by Google's numbers, uses 24% less quota. For teams on 3.8 Flash, it is an upgrade available on the app, Search and API on the same day.
Version History
- Release (2026-11-12): Google released Gemini 4 Hydrogen, the third model in the Gemini 4 line, under the API name gemini-4-hydrogen. Four thinking levels (low, medium, high by default, xhigh); native 1M-token input with up to 1M output tokens; $1.50 input and $7.50 output per million tokens, flat across the window. At the default level it finishes tasks 2.5x faster than Gemini 3.8 Flash and raises Terminal-Bench 4.1 from 18.2% to 57%. Same-day availability across the Gemini app, Search, Gemini API, AI Studio, Gemini for business, Antigravity and Sheets. Antigravity's Orchestrator can switch between Carbon and Hydrogen within a session, using 24% less quota and finishing 19% faster on Terminal-Bench 4.0.