Over the last few weeks, my main takeaway with the Fable 5 and GPT-5.6 releases was their respective costs and cost performances. Neither company particularly hides the fact that the real models they offer are paywalled. It creates a distinct entry barrier for those who simply are not the target audience of the marketing materials surrounding the AI hype: vibe coding, content creation, and so on.

Gemini, in particular, sits at an interesting spot on the spectrum. It is not stuck at voice assistant level, such as the current version of Siri, nor does it wall itself off behind professional and enterprise tiers. The Gemini Flash model, from my experience, is a good all-around LLM for everyday tasks. Compared to Claude, it often hallucinates with false positives, falsely thinking it found a hit, whereas Claude often hallucinates with false negatives, concluding the thing does not exist at all.

Synthetic benchmarks published since the release do back up where the community often places Gemini. It is the fastest of the big three. Free access is reasonably big for daily usage. All the while maintaining similar performance to other all-around models in its class. The new version this time focused primarily on speed and token efficiency. Compared to Sonnet 5, Anthropic’s recently released everyday model, the difference stems from respective positioning of the models. Sonnet 5 is available on free tier, but it runs out fast. Many Claude users also point out Opus overtakes Sonnet on cost performance, if the speed is taken out of the criteria.

My rule of thumb with all-around models is to never trust one model. If Sonnet says something is not possible, assume Gemini will say otherwise. If Gemini says something is possible, assume Sonnet to disagree. If Sonnet says something is not possible based on deduction, assume Gemini will disagree based on search results. My take with Gemini Flash is that it truly offers what most users want from a free model. For paying users, it’s better to ask a higher model expecting a better result, but for free users, being able to ask more at all is the higher priority.