Independent · Updated November 20, 2026 · Every claim traceable
Home › AI Models › Gemini 4 Neon: Benchmarks, Pricing & Antigravity Gui…
AI Models

Gemini 4 Neon: Benchmarks, Pricing & Antigravity Guide

By Yogesh · · 8 min read

Sources checked and dated on publication. See our editorial policy.

Google DeepMind introduced Gemini 4 Neon on November 18, 2026, slotting a mid-priced workhorse between Gemini 4…

Google DeepMind introduced Gemini 4 Neon on November 18, 2026, slotting a mid-priced workhorse between Gemini 4 Hydrogen and Gemini 4 Carbon at $3 per million input tokens and $15 per million output tokens. With a 1 million-token context window, four tunable thinking levels, and Orchestrator pairing in Google Antigravity, Neon targets teams running high-frequency coding, analytical, and cybersecurity workflows.

The short answer

Pricing and how Neon compares on token math

Gemini 4 Neon is priced at $3 per million input tokens and $15 per million output tokens, with cached input tokens priced at $0.15 per million (a 95% caching discount). On headline list prices, Neon sits 50% above Anthropic's mid-range Claude Sonnet 5.5 ($2 / $10) and OpenAI's mid-range GPT-6.2 Sol ($2 / $10), and twice the rate of Google's own Gemini 4 Hydrogen ($1.50 / $7.50).

Two pricing details narrow or reverse that gap in real workloads. First, Neon's list rates sit 25% below Anthropic's flagship Claude Opus 5.5 ($4 / $20) and 25% below Gemini 4 Carbon's scheduled standard rates ($4 / $20), which take effect on December 1, 2026 after Carbon's $2 / $10 introductory period ends on November 30. Second, under OpenAI's long-context tiering, GPT-6.2 Sol requests exceeding 272,000 input tokens jump to $4 input and $15 output for the entire request, erasing Sol's headline output-price advantage on large repositories or multi-document bundles.

Cost per task also depends heavily on how many reasoning tokens a model burns before producing an answer. Google reports that at its default High thinking level, Neon delivers a 32% lower cost per task than Gemini 4 Carbon on the Artificial Analysis Intelligence Index, and roughly 50% to 80% lower cost per task than top competing frontier models on the Vals Index and Agent Arena. Even at its Extra High (xhigh) setting, Google states Neon costs 36% or more less per task than competing models at their highest effort setting.

Coding, agentic, and enterprise benchmarks

Google's launch numbers emphasize multi-step tool use and repository-scale software engineering rather than isolated single-turn trivia. On the Artificial Analysis Intelligence Index, Neon scores 61 at its xhigh thinking level, placing #3 overall (Artificial Analysis confirmed the third-place ranking publicly). On the Artificial Analysis Coding Agent Index, Neon ranks #1 overall with a score of 58.6, alongside 65.2% on Terminal-Bench 4.1 and 46.2% at its default High setting on FrontierCode 2.0.

The harness matters noticeably on software engineering tasks. On SWE-sweep, Neon leads on the mini-SWE harness at 12.8%, and more than triples to 39.0% inside Google Antigravity. In enterprise and analytical evaluations, Google reports:

Independent reporting from VentureBeat notes an important caveat for buyers comparing vendor tables: Google's own head-to-head charts show that while Neon outscored GPT-6.2 Sol on business automation, Sol read dense professional charts more accurately. Furthermore, Google did not publish absolute wall-clock completion times for matched Neon versus Carbon workloads; the clearest speed comparison comes from early partner Glean, whose AI product lead Thai Tran reported matching or exceeding Carbon quality at 3x the speed and 60% of the cost in internal evaluations.

Antigravity Orchestrator: pairing Neon with Hydrogen

One of the most practical engineering patterns in this release is not a single-model benchmark, but how Google splits work between two models inside Google Antigravity. Alongside Neon's launch, Antigravity's Orchestrator mode now pairs Gemini 4 Neon as the planner with Gemini 4 Hydrogen as the executor inside one prompt and session.

In that loop, Neon handles upfront architecture, complex debugging, and multi-file planning, then delegates routine code edits, package updates, and test runs to the faster, cheaper Hydrogen model. When Hydrogen encounters unexpected test failures or state mismatches, Orchestrator automatically hands control back to Neon to inspect state, repair the plan, and resume execution. For teams watching monthly token budgets across continuous coding agents, this planner-executor split avoids paying Neon rates for boilerplate edits while preserving Neon's judgment when the plan breaks.

Developers using GitHub Copilot also gain immediate access. GitHub announced general availability of Gemini 4 Neon on November 18 for Copilot Pro+, Max, Business, and Enterprise plans across VS Code, Visual Studio, Copilot CLI, JetBrains IDEs, Xcode, Eclipse, github.com, and GitHub Mobile, billed at provider list pricing under usage-based billing.

Cyber defense and agentic safety controls

Google DeepMind paired Neon's release with dedicated cybersecurity benchmarks and a Fairwind Program rollout for eligible cyber defenders. On AutoBaxBench, Google reports Neon reaches 76% on the hard tier for generating secure code. Across vulnerability discovery, validation, and patching, Google reports 90.5% on SEC-bench Pro at roughly half the cost per challenge of competing large models, 51.26% on WebExploitBench v1.2, and 70.7% on Collinear's CWE-bench v2 (a 3.4% uplift over Carbon), plus frontier ExploitGym validation performance using half the tokens of Carbon.

For production agent deployments, reliability against prompt injection and shortcut-taking matters as much as raw intelligence. Google reports that with real-time safeguards enabled at its highest thinking level, Neon recorded a 0.0% attack success rate on Gray Swan's Indirect Prompt Injection (IPI) benchmark. On internal alignment evaluations, Google cites 0.0% cheating on Terminal-Bench 4.0, a 2.45% shortcut rate on its coding cheating evaluation (its lowest across Gemini 4 models), a 0.09% misaligned action rate on its agentic safety suite, and a 99.2% pass rate on agent honesty, while keeping over-refusals low at 1.98% on Terminal-Bench 4.0 and 0% on Terminal-Bench Science.

When to choose Neon over Hydrogen, Carbon, or rivals

For teams standardizing on the Gemini 4 stack, the decision tree is straightforward once December pricing kicks in. Keep Gemini 4 Hydrogen ($1.50 / $7.50) for high-volume subagent execution, routine edits, and latency-sensitive search or triage. Use Gemini 4 Neon ($3 / $15) as the default daily driver for repository-wide coding, multi-step enterprise analysis across PDFs and spreadsheets, and orchestrating Hydrogen subagents. Reserve Gemini 4 Carbon ($4 / $20 starting December 1) for the small fraction of deep, long-horizon research tasks where maximum single-model depth outweighs speed and per-task cost.

Against external rivals, Neon's strongest angle is predictable long-context pricing and steerability. If your agent workflows routinely cross 272K input tokens or rely on mid-task human corrections, Neon avoids the long-context pricing jump on GPT-6.2 Sol and undercuts Claude Opus 5.5 by 25% on list token rates while staying competitive across coding and enterprise benchmarks.

Frequently asked questions

Pricing and context window

Gemini 4 Neon costs $3 per million input tokens ($0.15 cached input, a 95% discount) and $15 per million output tokens across a 1 million-token context window.

Benchmark standings

Google reports a score of 61 (#3 overall) on the Artificial Analysis Intelligence Index at xhigh thinking, #1 overall on the Artificial Analysis Coding Agent Index (58.6), #1 on Zapier AutomationBench (56.3%), and #2 on the Vals Index (70.9%).

Antigravity Orchestrator pairing

Inside Google Antigravity, Orchestrator pairs Gemini 4 Neon as the planner with Gemini 4 Hydrogen as the fast executor, automatically handing back to Neon when Hydrogen hits friction.

Where it is available

Rolling out across the paid Gemini API in Google AI Studio and Android Studio, Gemini for business, GitHub Copilot (Pro+, Max, Business, Enterprise), and Google Antigravity and the Gemini app for Pro and Ultra subscribers.

Sources

Prices and benchmark figures are standard published rates as of November 20, 2026 and can change without notice. Some outbound links may be affiliate links — see our disclosure. Found an error? Tell us and we'll correct it in public.