Episode context
Gemini 3 Pro shipped with a lot of hype, so we ran our standard battery: planet generation, challenge-mode gameplay, a business reasoning workflow, and a broader Vibe Bench comparison to see how it stacks up in 3D, coding, and analysis.
Summary
The first planet pass came back fast and looked clean in atmosphere, but the clouds and biomes were underwhelming. A second pass improved water and terrain detail, yet the clouds and shader issues remained uneven. In challenge mode, Gemini 3 Pro surprised us with a solid first-person loop and better directional coherence than prior models, even if scale and UI bugs still showed up.
On the business reasoning test, it did a decent job assembling benchmarks and produced thoughtful charts, including a useful “commodity vs. frontier” framing. Recommendations were mostly reasonable, though we still saw stronger business reasoning elsewhere. Vibe Bench comparisons showed Gemini 3 Pro to be reliable in coding and strong in illustration, but often simpler or less imaginative in one-shot 3D scenes relative to the best frontier peers.
Key takeaways
- Gemini 3 Pro improves the first-person gameplay loop and directional coherence in challenge mode.
- Planet generation still struggles with clouds and biome diversity, even after feedback.
- Business reasoning output is solid and structured, with a clear value map, but not the best we’ve seen.
- Vibe Bench comparisons suggest Gemini 3 Pro is dependable but often less creative in one-shot 3D scenes.
- Overall performance feels like a step forward, not a sudden leap, reinforcing the “spiky intelligence” reality.
Highlight moments
- 00:24 — Planet gen kicks off with atmosphere, water, and biome focus
- 02:43 — Realism pass improves water and terrain but clouds remain weak
- 04:22 — Challenge mode lands a strong first-person loop
- 08:49 — Business reasoning charts and “commodity vs. frontier” framing
- 18:16 — Illustration comparison shows Gemini 3 Pro’s strengths outside 3D
Scorecard
Coming soon.
In our words
Gemini 3 Pro does what we expected: it moves the needle forward, especially in gameplay structure and coding reliability, but it isn’t a revolution. The best result was the challenge-mode loop, where coherence and directionality improved noticeably. In 3D one-shot scenes and business reasoning, it’s competitive but not dominant. This is a strong model with clear strengths and clear limits, and it reinforces the point that picking the right model per task matters more than crowning a single winner.