Uploaded November 2025 | Updated September 2026, 4 days ago
Gemini 3 is undeniably fast and impressive on benchmarks, but after a full week of real-world software engineering work, the reality is more complicated. While everyone's been hyping its capabilities based on day-one reviews and marketing materials, this video digs into what actually matters: how Gemini 3 performs with coding agents on real projects, not just one-shot Tetris games or simple websites. The speed is remarkable at 128 tokens per second, but it comes with serious trade-offs that affect daily pair programming work.
The core issues are frustrating: Gemini 3 is nearly impossible to redirect once it commits to a plan, suffers from an 88% hallucination rate (nearly double Sonnet 4.5's 48%), and confidently claims tasks are complete when they're not. It ignores context from earlier in conversations, struggles with complex multi-step instructions, and dismisses suggestions like a grumpy coder who thinks they know best. While it excels at one-shot code generation, it falls short as a collaborative partner for serious software development. Gemini 3 is genuinely one of the best models available (probably second place behind Sonnet 4.5) but it's not the massive leap forward that the hype suggests, and the gap between Claude Code and Gemini CLI remains significant.
#Gemini3 #AIcoding #SoftwareEngineering
Consider joining the channel: youtube.com/c/devopstoolkit/join
▬▬▬▬▬▬ 🔗 Additional Info 🔗 ▬▬▬▬▬▬
➡ Transcript and commands: https://devopstoolkit.live/ai/gemini-3-is-fast-but-gaslights-you-at-128-tokens-second
🔗 Gemini 3: https://deepmind.google/models/gemini
▬▬▬▬▬▬ 💰 Sponsorships 💰 ▬▬▬▬▬▬
If you are interested in sponsoring this channel, please visit https://devopstoolkit.live/sponsor for more information. Alternatively, feel free to contact me over Twitter or LinkedIn (see below).
▬▬▬▬▬▬ 👋 Contact me 👋 ▬▬▬▬▬▬
➡ BlueSky: https://vfarcic.bsky.social
➡ LinkedIn: linkedin.com/in/viktorfarcic
▬▬▬▬▬▬ 🚀 Other Channels 🚀 ▬▬▬▬▬▬
🎤 Podcast: devopsparadox.com
💬 Live streams: youtube.com/c/DevOpsParadox
▬▬▬▬▬▬ ⏱ Timecodes ⏱ ▬▬▬▬▬▬
00:00 Gamini 3 with Gamini CLI
00:25 Gemini 3 Real-World Testing
02:54 Gemini 3's Biggest Problems
10:10 Is Gemini 3 Worth It?
Gemini 3 is undeniably fast and impressive on benchmarks, but after a full week of real-world software engineering work, the reality is more complicated. While everyone's been hyping its capabilities based on day-one reviews and marketing materials, this video digs into what actually matters: how Gemini 3 performs with coding agents on real projects, not just one-shot Tetris games or simple websites. The speed is remarkable at 128 tokens per second, but it comes with serious trade-offs that affect daily pair programming work.
The core issues are frustrating: Gemini 3 is nearly impossible to redirect once it commits to a plan, suffers from an 88% hallucination rate (nearly double Sonnet 4.5's 48%), and confidently claims tasks are complete when they're not. It ignores context from earlier in conversations, struggles with complex multi-step instructions, and dismisses suggestions like a grumpy coder who thinks they know best. While it excels at one-shot code generation, it falls short as a collaborative partner for serious software development. Gemini 3 is genuinely one of the best models available (probably second place behind Sonnet 4.5) but it's not the massive leap forward that the hype suggests, and the gap between Claude Code and Gemini CLI remains significant.
#Gemini3 #AIcoding #SoftwareEngineering
Consider joining the channel: youtube.com/c/devopstoolkit/join
▬▬▬▬▬▬ 🔗 Additional Info 🔗 ▬▬▬▬▬▬
➡ Transcript and commands: https://devopstoolkit.live/ai/gemini-3-is-fast-but-gaslights-you-at-128-tokens-second
🔗 Gemini 3: https://deepmind.google/models/gemini
▬▬▬▬▬▬ 💰 Sponsorships 💰 ▬▬▬▬▬▬
If you are interested in sponsoring this channel, please visit https://devopstoolkit.live/sponsor for more information. Alternatively, feel free to contact me over Twitter or LinkedIn (see below).
▬▬▬▬▬▬ 👋 Contact me 👋 ▬▬▬▬▬▬
➡ BlueSky: https://vfarcic.bsky.social
➡ LinkedIn: linkedin.com/in/viktorfarcic
▬▬▬▬▬▬ 🚀 Other Channels 🚀 ▬▬▬▬▬▬
🎤 Podcast: devopsparadox.com
💬 Live streams: youtube.com/c/DevOpsParadox
▬▬▬▬▬▬ ⏱ Timecodes ⏱ ▬▬▬▬▬▬
00:00 Gamini 3 with Gamini CLI
00:25 Gemini 3 Real-World Testing
02:54 Gemini 3's Biggest Problems
10:10 Is Gemini 3 Worth It?
