Skip to content

AI · 20 November 2026 · 5 min read

Google DeepMind Releases Gemini 4 Neon With Unified 1M Context and Lower Agent Task Costs

Priced at $3 input and $15 output per million tokens, Google’s fourth Gemini 4 model targets long-context knowledge work, though a temporary Carbon discount complicates November adoption.

A white outdoor sign with the Google name in its usual colours, set among trees.
A Reuters photograph of a Google sign. It is not DeepMind’s office, and it is not the Gemini 4 Neon model.

Lede & Nut Graf

Google DeepMind released Gemini 4 Neon (`gemini-4-neon`) on November 18, 2026—its fourth Gemini 4 family release in under two months—positioning the model as a mid-tier workhorse optimized for lower total cost per completed task (Google Blog, Nov. 18, 2026; Google AI Studio documentation, Nov. 20, 2026). Available exclusively on the paid tier of the Gemini Application Programming Interface (API) and to Google AI Pro and Ultra subscribers, Neon targets enterprise agent workflows that bog down in multi-turn tool calls.

For enterprise AI engineering teams, the release underscores a shift in inference economics. As reasoning models burn millions of hidden "thinking tokens," per-token list prices matter far less than how efficiently a model resolves a task without redundant file reads or context bloat.

Technical & Financial Breakdown

Gemini 4 Neon uses a unified 1,048,576-token context window shared across input, thinking, and output tokens, supporting synchronous outputs up to 65,536 tokens and long decode up to 1 million tokens via background execution or client-side continuation (Google AI Studio docs, Nov. 20, 2026). Developers can tune reasoning depth across four levels—`low`, `medium`, `high` (default), and `xhigh`—while standard sampling knobs (`temperature`, `top_p`, `top_k`) are ignored or rejected.

Flat standard API pricing sits at $3.00 per million input tokens and $15.00 per million output tokens (including thinking tokens), with Batch and Flex tiers priced at $1.50/$7.50 and Priority at $5.40/$27.00 (DataCamp, Nov. 19, 2026). Explicit context caching is unsupported; instead, implicit caching triggers automatically on prompts above 4,096 tokens, discounting cached input by 95% to $0.15 per million tokens.

On vendor-reported and independent benchmarks, Neon leads on long-context and knowledge work while trailing slightly on pure terminal coding (DataCamp, Nov. 19, 2026):

  • Long Context (GraphWalks, 256K–1M): 95.2% vs. 80.1% for GPT-6.2 Sol and 66.8% for Claude Opus 5.5.
  • Knowledge Work (AutomationBench): 56.3% vs. 42.5% for Opus 5.5 and 41.3% for GPT-6.2 Sol.
  • Agentic Coding: Leads CursorBench 4.1 (65.4% vs. 58.1% for Opus 5.5), but trails Opus 5.5 on Terminal-bench 4.1 (65.2% vs. 65.8%) and FrontierCode 2.0 (46.2% vs. 47.3%).
  • Efficiency: Google reports 32% lower cost per task than Gemini 4 Carbon at default `high` reasoning and an Artificial Analysis Intelligence Index score of 61 at `xhigh`.

Second-Order Implications

With Neon slotted cleanly between Gemini 4 Hydrogen ($1.50/$7.50) and Gemini 4 Carbon (standard $4.00/$20.00), Google is building a tightly tiered inference ladder designed to keep enterprise workloads inside its custom Tensor Processing Unit (TPU) ecosystem. The 95.2% GraphWalks score in particular gives Google a tangible engineering edge for legal, financial, and codebase-wide retrieval workloads where rival models still degrade past 256,000 tokens.

Caveats & Open Questions

Neon’s launch timing creates a short-term pricing anomaly: Gemini 4 Carbon’s promotional rate of $2.00/$10.00 runs through November 30, 2026, making Neon 50% more expensive per token than Google’s own flagship until Carbon resets to $4.00/$20.00 on December 1 (DataCamp, Nov. 19, 2026). In addition, Google’s claims that Neon reduces redundant tool calls and context-compaction errors remain vendor assertions that engineering teams must validate against their own production agent logs ``.