SYLIN

Gemini Hydrogen

by Google · Model · Language & code models

mixed

Fast and cheap for everyday agent work, but behind rivals on several tests.

Powers free Gemini Tasks in six countries, including tasks needing a phone call or text

AccessAPIPrice$1.5 in · $7.5 out per M tokensContext1,049K tokens

Why we say so

Artificial Analysis scores it one point behind Sonnet 5.5 and GPT-6.1 Sol on its index, with a far lower hallucination rate (8%), and Vals ranks it sixth. Terminal-Bench is below Sonnet and Sol. Simon Willison found its output starkly worse than Carbon's for only a 25% discount, and one tester saw it burn 131K thinking tokens without answering. Neon beat it on CursorBench (63.7% vs 57.3%).

Evidence

  1. Nov 19
    Antigravity multi-model mode: Carbon plans, Hydrogen codes; near-Carbon Terminal-Bench 4.0 with 24% less quota, 19% faster; Ultra rollout
  2. Nov 15
    Fast Gemini 4 workhorse for everyday coding and agentic work; text, image, audio, video and file input, 1M-token context, adjustable thinking levels
  3. Nov 14
    Roboflow vision bench at high effort: 29.3s per image vs 51.4s Argon and 132.3s Carbon, though Carbon detects more accurately
  4. Nov 14
    Edges Argon on Artificial Analysis Intelligence Index: 51 vs 50 at default thinking, 54 at xhigh
  5. Nov 13
    Google releases fast Gemini 4 tier: 1M context, four thinking levels, $1.50/$7.50 per million tokens, free tier in AI Studio
Show all 11
  1. Nov 13
    Google says it completes tasks 2.5x faster than Gemini 3.8 Flash
  2. Nov 13
    At xhigh scores 54 on AA Intelligence Index, one point behind Sonnet 5.5 and GPT-6.1 Sol; about 32% of Sonnet's cost, 2.5x Sol's
  3. Nov 12
    About 15x the price of Claude Haiku 5.5 and GPT-6 Luna, which are both $0.10/$0.50 per million tokens
  4. Nov 12
    Google pitches it against Meta's Muse Ember and likely Claude Haiku 5.5 at its price tier
  5. Nov 12
    Google's table has it ahead of Sonnet 5.5 and GPT 6.1 Terra on most rows: bug fixing, knowledge work, charts, hard reasoning
  6. Nov 12
    Claimed score-versus-time chart: at top setting matches Sonnet's best score in well under half the time

Cost to run

Alternatives

Gemini CarbonOn long complex tasks it did better than Hydrogen in similar time at about 50% more cost; on simple tasks too slow and costly
Gemini NeonScored 63.7% on CursorBench 4.1 at default effort, up from Hydrogen's 57.3%