Gemini 4 Hydrogen replaces 3.8 Flash at the same $1.50 in and $7.50 out per million tokens, with four thinking levels to price each step
Google just put a near frontier model at Flash prices, and the real decision is now which thinking level each step of a workflow gets.
What happened
Google released Gemini 4 Hydrogen, the replacement for Gemini 3.8 Flash across its products, according to Ars Technica. API pricing stays at $1.50 per million input tokens and $7.50 per million output tokens, the same as 3.8 Flash, and The Decoder reports cache hits at $0.075. It has a one million token context window for input and can output up to one million tokens.
Hydrogen runs at four thinking levels: low, medium, high (the default) and a new xhigh. Google says it finishes tasks 2.5 times faster than 3.8 Flash at the default level, cuts hallucinations by a factor of five, and recovers better when a tool call fails or a search comes up empty. On Terminal-Bench 4.1, Google puts it at 57 percent versus 18.2 percent for Flash.
The independent numbers from Artificial Analysis, as reported by The Decoder, are the useful part: about $3.16 per task at xhigh, $1.87 at high, $0.80 at medium and $0.29 at low. Its index score falls from 54 at xhigh to 51 at high, 41 at medium and 27 at low. It is weaker on agentic terminal work than Claude Sonnet 5.5 and GPT-6.1 Sol. Ars notes Carbon and Argon go from $2/$10 to $4/$20 next month.
My take
For most business automations, this is the model I would test first for the bulk work: classifying inbound leads, summarizing calls, pulling fields from documents, drafting follow ups. The price did not move, so swapping it into an existing Flash workflow is mostly a regression test.
The thing to plan around is the thinking level. The gap between low and xhigh is more than 10x in cost per task, and the quality gap is just as wide. So do not set one level for the whole workflow. Run the routing and tagging steps on low or medium, and save high for the step where a wrong answer costs money, like a quote or a CRM update.
Two cautions. Benchmarks are not your data, so run 30 to 50 real records from your client through it before switching. And if you built anything on Carbon or Argon at today's price, re-cost it now, before December's increase lands on the invoice.
More posts
- LangSmith can now hide trace inputs and outputs from engineers while still showing them latency, cost and run treesNov 19, 2026
- Microsoft-Decision-1.1 is generally available for routing and classification at $0.042 per million input tokens, with free outputNov 19, 2026
- LangSmith now ties coding agent spend to pull requests, so teams can see cost per merged PR across Claude Code, Codex and CursorNov 19, 2026
