Google launches Gemini 4 Hydrogen

Google DeepMind announced Gemini 4 Hydrogen, the execution-focused member of the Gemini 4 family. Its frontier models, Gemini 4 Carbon and Gemini 4 Argon, remain the choice for deep, long-horizon reasoning. The launch was rumoured earlier by our rumour desk.

On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default "high" thinking level, against Argon's 50. It reaches 54 at a new "xhigh" setting. Artificial Analysis measured 41 at medium and 27 at low, where an index task takes 1.3 minutes (13.0 at xhigh) and costs $0.29, with costs varying up to 11x across settings. Low also has the fastest time to first answer token, 5.81s; output speeds run 177–236 tokens/s. Pricing is the same across its 1 million-token context window: $1.50 per million input tokens and $7.50 per million output tokens. It can also produce up to 1 million output tokens.

Google pitches Hydrogen as the upgrade from Gemini 3.8 Flash, saying it finishes tasks 2.5x faster at its default setting, scores 57% on Terminal-Bench 4.1 versus Flash's 18.2%, and makes fabricated claims more than 5x less often. In early partner testing on graph, SQL and data-heavy enterprise workflows, Hydrogen ran 2x to 2.8x faster than 3.8 Flash and used 12% to 23% fewer tokens, Google says. Other Google figures:

• 58.8% on HLE-Diamond
• 52.3% on Vals Tax Agent Bench, ranked #1
• 55.8% on Vals Legal Research Bench, tied for #1 (Vals' leaderboard has 53.85%, #6 of 94)
• 53.4% on the new Multimodal Knowledge Agent benchmark
• 71.6% on Agents' Last Exam v1.1
• #5 on arena.ai's Agent Arena

Google also introduced Orchestrator in Google Antigravity, which switches between Carbon and Hydrogen within a session. On Terminal-Bench 4.0, Google says it neared Carbon's success rate using 24% less quota and finishing 19% faster.

Vals AI's testing at xhigh confirms Hydrogen at #1 of 85 on Tax Agent Bench (52.30%), ahead of Carbon (51.78%) and Claude Fable 5.1 (50.75%). It is #2 of 40 on the Vals Finance Index (66.14%, behind Carbon's 67.05%), #6 of 37 on the Vals Index (68.49% at $9.84 per test, behind Claude Fable 5.2's 71.31%) and #6 of 42 on EMB v1.1 (75.80%). On coding it is mid-pack: #11 of 64 on Terminal-Bench 4.0 (54.04%) and #13 of 128 on Vibe Code Bench (87.78%).

Hydrogen is available in Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and to AI Pro and Ultra subscribers in Antigravity.