Google has integrated Gemini 4 Neon into GitHub Copilot, marking a significant upgrade for developers using the tool across IDEs and platforms. The model is now generally available for Copilot Pro+, Max, Business, and Enterprise users, delivering faster response times, tighter code edits, and improved reliability in complex workflows. According to Google, Neon achieves quality close to its more powerful Gemini 4 Carbon while reducing token costs and minimising unrequested changes.
A balance of speed, cost, and intelligence
Gemini 4 Neon is engineered for efficiency without sacrificing performance. With a 1 million-token context window and pricing at $3 per million input tokens ($0.15 per million cached input tokens) and $15 per million output tokens, it undercuts larger models on cost per task. At its default high thinking level, Google claims a 32% reduction in cost per task compared to Carbon, and up to 80% lower than top competing frontier models on benchmarks like Vals and Agent Arena. Independent analysis by micjm.com suggests Neon’s projected Elo score on the Agent Duel leaderboard (1,066) places it just below Opus 5.5 (1,078), with speed playing a critical role in real-world performance.
Neon processes input at up to 127 tokens per second via OpenRouter—nearly three times faster than Carbon’s 46 tps. This speed advantage makes it particularly effective for agentic workflows, where rapid tool use and multi-step reasoning are essential. In internal testing, Neon shows a 0.09% misaligned action rate and 99.2% pass rate on honesty evaluations, suggesting tighter control over autonomous behaviour.
Strong performance across enterprise and coding tasks
Neon excels in complex knowledge work, ranking #2 on the Vals Index (70.9%) and achieving top scores on legal, tax, finance, and spreadsheet modelling benchmarks. It leads on Zapier’s AutomationBench (56.3%) and Andon Labs’ Vending-Bench 2 ($16,843), simulating long-horizon business operations. In scientific reasoning, it ranks first on Terminal-Bench Science (70.5%), RiemannBench (84.0%), and HLE-Diamond (71.3%).
For developers, Neon ranks #1 on the Artificial Analysis Coding Agent Index (58.6) and achieves 65.2% on Terminal-Bench 4.1. On SWE-sweep, it reaches 39.0% in Google Antigravity—a more than threefold improvement over its 12.8% in mini-SWE—demonstrating the impact of advanced tooling and orchestration. It handles multi-file debugging and refactoring with surgical precision, reducing redundant file reads and managing context compaction effectively.
Cybersecurity and safety by design
A specialised version, Gemini 4 Neon for cyber, is rolling out to trusted defenders via Google’s Fairwind Program. It sets new benchmarks in secure code generation, achieving 76% on AutoBaxBench hard challenges. On WebExploitBench v1.2, it reaches 51.26%—state-of-the-art at half the cost of rivals. It also leads in vulnerability detection (90.5% on SEC-bench Pro), exploit validation (using half the tokens of Carbon), and patching (70.7% on CWE-bench v2), improving over Carbon by 3.4 percentage points.
Safety mechanisms are tightly integrated. Neon matches Carbon’s protections against prompt injection and misuse, with real-time safeguards resisting all attacks in Gray Swan’s IPI benchmark. It refuses only 1.98% of tasks on Terminal-Bench 4.0 and 0% on science tasks, suggesting high tolerance for legitimate work. Misaligned actions dropped to 2.45% in coding evaluations—the lowest in the Gemini 4 lineup.
What it means
For UK, UAE, US, and Kazakh businesses, Neon’s integration into Copilot offers a compelling alternative to Opus-class models: faster, cheaper, and more controllable. Enterprises relying on long-running agent workflows can reduce latency and cost while maintaining high output quality. The Fairwind rollout signals Google’s focus on secure, responsible deployment in high-stakes domains.
Developers should test Neon in multi-step coding tasks, especially where cost efficiency and speed matter. Organisations using Copilot at scale can expect lower operational costs and tighter governance. However, access remains limited to higher-tier plans, excluding free and standard users.
Organisations should assess Neon’s performance in their specific workflows—particularly in legal, financial, and security contexts—while monitoring Google’s roadmap for broader availability.