What Is Gemini 3.1 Pro? Google's Thinking Model
What Is Gemini 3.1 Pro?
Gemini 3.1 Pro is Google DeepMind’s latest reasoning model, released on February 19, 2026. Its headline feature is three adjustable thinking levels — LOW, MEDIUM, and HIGH — that let you control exactly how much computational reasoning the model applies to each request.
Last verified: March 2026
Key Specifications
| Feature | Detail |
|---|---|
| Developer | Google DeepMind |
| Released | February 19, 2026 |
| Context window | 1M tokens |
| Thinking levels | LOW / MEDIUM / HIGH |
| Multimodal | Text, images, video, audio, code |
| API | Google AI Studio, Vertex AI |
Adjustable Thinking Levels
The standout feature. Instead of one-size-fits-all reasoning, Gemini 3.1 Pro lets you pick the depth:
LOW — Quick Responses
- Simple questions, factual lookups, translations
- Fastest response time, lowest cost
- Comparable to Gemini 3 Flash in speed
MEDIUM — Balanced Reasoning
- General coding tasks, writing, analysis
- Good balance of speed and accuracy
- Default mode for most interactions
HIGH — Deep Problem Solving
- Complex math, multi-step reasoning, scientific analysis
- Takes longer but applies thorough chain-of-thought
- Functions like a “Deep Think Mini” — not as powerful as the full Gemini Deep Think but significantly more capable than standard mode
This flexibility means you’re effectively getting three models in one. A quick question doesn’t burn the compute budget of a complex analysis.
What’s New vs. Gemini 3 Pro
| Capability | Gemini 3 Pro | Gemini 3.1 Pro |
|---|---|---|
| Thinking levels | No | YES (3 levels) |
| Reasoning | Good | Significantly improved |
| Coding | Strong | Stronger |
| Context | 1M | 1M |
| Speed | Fixed | Variable (by level) |
| Factual accuracy | Good | Improved |
The upgrade isn’t just the thinking levels. Google DeepMind improved the underlying model’s core intelligence — better problem decomposition, stronger logical reasoning, and improved factual accuracy across all levels.
Pricing
| Thinking Level | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| LOW | $0.50 | $1.50 |
| MEDIUM | $1.25 | $5.00 |
| HIGH | $2.50 | $10.00 |
The tiered pricing means you only pay for deep reasoning when you need it. For high-volume applications that mix simple and complex queries, this can significantly reduce costs compared to models with fixed pricing.
Best Use Cases
- Coding assistants — Use LOW for completions, HIGH for architecture decisions
- Customer support — LOW for FAQs, MEDIUM for troubleshooting, HIGH for escalations
- Research — HIGH for analysis, LOW for summarization
- Content creation — MEDIUM for drafting, HIGH for technical accuracy
How to Access
Google AI Studio — Free tier available with rate limits. Full access through paid plans.
Vertex AI — Enterprise access with SLAs, VPC integration, and data residency controls.
API — Set thinking level per request via the thinkingLevel parameter:
thinkingLevel: "HIGH"Competitive Position
Gemini 3.1 Pro sits between GPT-5.4 and Claude Opus 4.6 in overall capability, but the adjustable thinking levels give it unique flexibility. No other major model offers this level of per-request reasoning control. It’s particularly cost-effective for applications that need variable reasoning depth.
Last verified: March 2026