gemini-4-hydrogen is live: default high thinking and 57% on Terminal-Bench 4.1
Google’s Gemini 4 speed tier, released on 12 November and adapted from Carbon, is positioned as the upgrade path from 3.8 Flash. APIYI connected on 14 November; it is available in the default and svip groups on both native and OpenAI-compatible endpoints.
Pricing per 1M tokens: $1.50 input, $7.50 output (including thinking), $0.075 cached input. The rate is flat across the full 1M window with no expiry.
Migration notes: thinking defaults to high with four levels; minimal is rejected and thinkingBudget is deprecated; sampling parameters are ignored; keep max_tokens at 1024 or higher because thinking counts against it; return thoughtSignature verbatim in multi-turn and tool calls.
Read the full analysis in Gemini 4 Hydrogen Is Live and see Model Pricing.
← Back to Live Updates · 📚 Monthly archive