AI
Creator’s Gemini 4 testing finds strong niche abilities but serious coding reliability problems
Theo’s hands-on review portrays Gemini 4 Argan as useful for YouTube workflows, some vision tasks and selected frontend work, but unreliable for serious software engineering. His tests found low cash-hit rates, high costs after introductory pricing, unstable tooling and numerous false findings in codebase audits. He emphasizes that benchmark results depend heavily on task design and contrasts them with his own real-world evaluations, where Gemini produced oversized, incorrect or untested changes. The conclusions are strongly negative and remain one reviewer’s experience, not an independent industry assessment.
Editorial summary · Public video captions
What matters
- The reviewer says introductory API pricing rises from $2/$10 to $4/$20 per million input/output tokens on December 1.
- In one code audit, he reports 24 false claims out of 36 from Gemini.
- He found Google’s unique YouTube access useful despite broader dissatisfaction with the model.
Context & caveats
- The source is a machine-transcribed video and contains uncertain names, numbers and wording; performance claims come from the creator’s own tests and opinions.
Why this made the edition
The complete video transcript contains extensive hands-on testing, comparative benchmarks, pricing analysis and concrete coding examples, while clearly representing one creator’s evaluation rather than independent consensus.
Source material