AI · 20 November 2026
Google Gemini 4 Neon: Cheaper AI Coding Agents for Complex Tasks
Google has launched Gemini 4 Neon, a model aimed at complex software engineering that the company says can complete work more economically than larger models. It is priced at $3 per million input tokens, according to reporting by InfoWorld.
What happened
Google is rolling out Gemini 4 Neon, a model built for complex software engineering that the company says can complete work more economically than larger models. The launch, reported by InfoWorld, is aimed at expanding the use of AI coding agents among enterprise developers.
According to Google, Neon can work across large code repositories, carrying out multi-file debugging and refactoring while making targeted changes to existing code. Within Google's Antigravity development environment, it can also orchestrate several Gemini 4 Hydrogen subagents working in parallel.
Google positions Neon between two other models in its Gemini 4 range: Gemini 4 Carbon, designed for demanding projects, and Gemini 4 Hydrogen, which prioritises faster execution. API pricing is set at $3 per million input tokens.
Why it matters
The significance lies less in any single benchmark and more in the shift it signals. AI coding tools are being pitched not just as assistants that write isolated snippets, but as agents that can navigate a whole codebase, change existing code with precision and divide work among parallel subagents. For engineering organisations, that moves the conversation from whether an agent can help to how much work it can absorb, and at what cost per task.
Tiering is also becoming a design question. By placing Neon between a deep-reasoning model and a speed-first model, Google is effectively asking teams to match the model to the job. That mirrors how service operations already route cases by complexity, and it means cost and latency will increasingly shape how digital teams organise their work.
By the numbers
- 58.6 score on the Artificial Analysis Coding Agent Index, where Google says Neon ranked first
- 65.2% achieved on Terminal-Bench 4.1, a benchmark for AI agents performing tasks in terminal environments, per Google
- $3 per million input tokens, the API price Google has set for Neon
The Renascence take
Leaderboard positions are Google's framing, and they tell leaders less than they appear to. The harder question is where agentic coding breaks down in practice, which is usually at review, handoff and accountability rather than at generation.
Benchmark rankings will not decide whether these agents earn their place in your engineering teams. What will decide it is whether developers trust the changes they are asked to review, and targeted edits to existing code are a better bet for that trust than sweeping rewrites. Treat the Neon, Hydrogen and Carbon tiers as a service design decision: map which tasks need depth, which need speed, and which can tolerate a cheaper model. Then measure review time and rework, not just cost per token, because a cheap change that takes longer to verify is not actually cheaper.
Sources
This briefing was written by our Newsdesk, synthesising reporting from the outlets below. Follow the links for the original coverage.
FAQ
Questions we get on this topic
More in AI
Stay ahead of CX
Get the signal, not the noise.
The stories shaping customer experience — plus the Journal and Experience Loom — in your inbox.


