deniz.in

Markets

Weather

Loading weather

· via Vercel blog

Google's Gemini 4 Neon coding model arrives on Vercel's AI Gateway

Vercel has added Google's Gemini 4 Neon to its AI Gateway, a model tuned for coding and agent workflows with a 1M-token context window and four reasoning levels.

Google's Gemini 4 Neon coding model arrives on Vercel's AI Gateway

Vercel has added Google's Gemini 4 Neon to its AI Gateway service, according to a changelog post published on November 18. The move gives developers on the platform a new option for the model categories that dominate current development work: coding assistants, autonomous agents and other multi-step pipelines.

A step up from the lighter Gemini 4

According to Vercel, Neon is a major upgrade over the earlier, lighter Gemini 4 model. The company says it delivers output quality close to Google's larger Gemini 4 models while responding faster, which places it between the two tiers: near-flagship results with better speed. Vercel describes the model as purpose-built for coding, agent workflows and multi-step tasks — the kind of work where a model is invoked many times in sequence rather than in a single exchange.

Multimodal input and a large context window

Neon accepts text, images, video and documents as input, Vercel says, and offers a context window of one million tokens. A context that size can absorb lengthy documents or large amounts of code, which matters in agentic setups where conversation history, tool results and relevant files accumulate quickly over a run.

Tool calling, grounding and reasoning controls

For agent builders, Vercel lists support for tool calling and structured outputs, along with grounding in Google Search and Google Maps for queries that benefit from live or location-aware data. The model also exposes four reasoning levels — low, medium, high and xhigh. The naming suggests a dial for how much deliberation the model performs before answering, a trade-off that typically pits response latency against answer depth.

How to access it

On AI Gateway, the model is available under the ID google/gemini-4-neon. Vercel points developers to its coding agents guide for wiring the model into Codex, Cursor or another supported agent, and says it can also be tried directly in the model playground alongside the other language models the gateway offers.

Why it matters

Coding and agent pipelines are sensitive to latency and quality at the same time: a model that gets called dozens of times per task has to be fast, but one weak answer can break the whole chain. If Neon delivers what Vercel claims — quality approaching Google's larger Gemini 4 models at quicker response times — it becomes a credible default for those workloads rather than a compromise choice. Its availability through AI Gateway also lowers the barrier to evaluating it: teams already routing requests through Vercel can swap in a new model ID, compare behaviour in the playground, and test it against alternatives without adding another vendor relationship. As with any vendor performance claims, the quality and speed figures come from Vercel and Google, so developers building production systems will want to benchmark Neon on their own tasks before committing to it.

  • #google
  • #gemini
  • #vercel
  • #ai-gateway
  • #coding-agents
  • #llm

Related posts