Google has released Gemini 3.1 Ultra, calling it its most significant model launch of the year. The headline feature is a 2-million token context window that works natively across text, image, audio, and video — without converting everything to text first.
What’s new in Gemini 3.1 Ultra
- 2-million token context window: You can feed in entire codebases, hours of video, or hundreds of documents in a single conversation.
- Native multimodal processing: Image, audio, and video are processed directly by the model rather than being transcribed into text first, which Google says improves accuracy on tasks like analyzing a video call or a scanned document with charts.
- Sandboxed Code Execution tool: Gemini can now write, run, and test code in a secure sandbox mid-conversation, then show you the actual output — useful for data analysis, debugging, and quick prototyping without leaving the chat.
- Improved grounding: Google says factual queries are significantly less likely to produce hallucinated answers thanks to better grounding with search and verified sources.
How it compares to GPT-5.5 and Claude Opus 4.8
Each of the three major labs shipped a flagship update within days of each other in June 2026. Broadly:
- Gemini 3.1 Ultra leads on raw context size (2M tokens) and native multimodal input.
- GPT-5.5 focuses on reduced hallucinations (52.5% fewer than its predecessor) and leads Terminal-Bench 2.0 at 82.7%.
- Claude Opus 4.8 leads on long, multi-step agentic work with parallel subagents and scores 88.6% on SWE-bench Verified.
In practice, Gemini 3.1 Ultra is the strongest choice when you need to process huge amounts of mixed media at once — for example, analyzing a long video plus its transcript plus related documents in one go.
What this means for you
- Everyday users: If you use the Gemini app, you may see Gemini 3.1 Ultra offered for complex tasks like analyzing long PDFs, spreadsheets, or videos you upload.
- Developers: The Code Execution tool means you can ask Gemini to not just write code, but actually run it and return real output — handy for quick scripts, data transformations, or testing a regex before you use it.
- Heavy context users: If you’ve hit context-length errors with other tools when pasting in large documents or codebases, Gemini 3.1 Ultra’s 2M token window is currently the largest available from a major provider.
FAQ
Is Gemini 3.1 Ultra free to use?
Ultra-tier models are typically available to paid Gemini Advanced subscribers and via the API at higher per-token pricing than Flash or Pro tiers.
Does the 2M context window cost more?
Yes — larger context windows generally come with higher per-token costs on the API, so it’s most cost-effective for tasks that genuinely need that much context.
Bottom line: Gemini 3.1 Ultra is built for scale — massive context windows and native multimodal understanding, with a built-in code sandbox that turns it into a lightweight dev environment.