Google's Gemini 4 Hydrogen lands on AI Gateway: stronger agentic coding and tool use, same latency class

Google's Gemini 4 Hydrogen can now be called through AI Gateway, succeeding the earlier Gemini 3.8 Flash and aimed at agentic coding, knowledge work and tool use. Gemini 3.8 Flash is the model it succeeds, and the new model improves those capabilities while staying in the same latency class. The release states the comparison as handling longer engineering tasks and producing better-researched documents, spreadsheets and presentations in fewer agent steps.

The model follows system instructions, formatting requirements and task boundaries more closely, with more reliable tool orchestration across long conversations. Improvements in visual reasoning and technical problem-solving support multimodal workflows, and stronger grounding reduces unsupported claims in search-backed answers. The same release says invented tool or test results in coding reports are reduced as a result; calls use the model name google/gemini-4-hydrogen.

The release describes the model's capability scope when called through AI Gateway, covering agentic coding, knowledge work, tool use and multimodal workflows; access includes selecting google/gemini-4-hydrogen in agents such as Claude Code, Codex and Cursor.

The capability claims come from the releasing party, with no benchmark scores or third-party evaluation attached; the same latency class and fewer agent steps are not given specific figures.