Gemini 4 Neon is a reason to re-test your document and agent pipelines, but not a reason to rebuild them. If you process contracts, invoices or tender files with an LLM, run 30-50 of your own real documents through Gemini 4 Neon alongside your current model and compare errors and cost before switching.
Google DeepMind released Neon on November 18. It costs $3 per million input tokens, $0.15 for cached input and $15 per million output tokens, with a 1 million-token context window. Google reports first place on Legal Research Bench, Tax Agent Bench and Zapier's AutomationBench. It also reports its lowest fabrication rate yet. The model is already available in the Gemini API and in Gemini for business.
What changes for a small business
The most practical detail is the 95% discount on cached input. Many pipelines send the same long instructions, checklist or contract template with every document, so a large share of the bill comes from repeated input. The 1M context also fits a full tender pack in a single call, which is the kind of job a tender analysis pipeline I built handles. One caveat: Google chose these benchmarks itself, and it measured fabrication internally. On AutomationBench, Neon's winning score is 56.3%, so it still failed almost half of the tasks. Any agent that writes to a CRM or sends invoices still needs a human check.
Why it matters: in a well-built pipeline, swapping the model is a config change rather than a new project. Pick the model that makes fewer mistakes on your own documents at a price you can live with.
Source: Google DeepMind