After the incremental improvement of Gemini 3.6 Flash, Google has struck again with Gemini 3.7 Flash, another incremental improvement. Apparently, it was supposed to be Gemini 3.5 Pro, but after it underperformed, DeepMind simply rebranded it.

## Model Specs & Pricing Let’s talk about its specs. It has a **1M token context window**, costs **$0.75/M input tokens** and **$3.75/M output tokens** (but it has a 50% off sale right now on OpenRouter). It scores a **56 on the Intelligence Index**, 3 points above DeepSeek V4 Pro 0813, but in Arena.ai’s leaderboard, it scores around the same. So, let’s run it through the usual benchmarks, plus some more, as I do think this is a really interesting model.

## Test 1: Photorealistic 3D Glass Sphere with Origami Crane We’ll be using Google AI Studio to try out this model. For the SVG origami crane generation, it outputted this self-contained vector illustration:

```html ```

**Evaluation:** It’s actually quite good, with the crane and glass looking really realistic. My only gripe is that the crane’s head is covered a little.

## Test 2: 3D Procedural Globe Generation Next up: the world generation benchmark featuring a full Three.js procedural globe with custom Simplex noise shaders and particle rings:

```html ```

**Evaluation:** Again, this is the most maximalist generation I’ve seen, which is really good. It proves the Gemini team is actually trying to make their model less lazy, which has been a really big problem since the start.

## Test 3: 3D Low-Poly Wizard Adventure Game Now for a harder prompt: > *"Create a fully functional, browser-based 3D third-person action-adventure game using Three.js inside a single self-contained HTML file, featuring a low-poly wizard exploring a vibrant floating island archipelago with bright pastel colors, dynamic day-night lighting, and environmental particle effects. Implement smooth WASD movement, spacebar jumping, and mouse-drag camera controls, complete with responsive collision detection, a dynamic camera-follow script, and an objective where the player collects glowing crystal artifacts while dodging automated elemental turrets. Include a clean user interface with a real-time score and health HUD, basic Web Audio API sound effects for jumping and collecting items, and a seamless game-over and restart system."*

```html ```

**Evaluation:** Even with this complex prompt, the model did incredibly well, and made a legitimately good, detailed game. This is the most maximalist generation I’ve seen, which is really good. It proves the Gemini team is actually trying to make their model less lazy, which has been a really big problem since the start.

## Final Verdict Overall, Gemini 3.7 Flash is a really great AI model that shows that Gemini is still in the frontier AI race. Hopefully, with a few new model releases, they’ll be back as the leaders of the pack. **Overall Score: 10 / 10**