An AI Practitioner's Daily Driver Is Shifting from Claude Opus 5.5 to Gemini Neon
A tech practitioner tested Gemini Neon and considers it the first Gemini model that can serve as a daily driver — so much so that he's thinking about replacing Claude Opus 5.5. Several research explainer videos he generated with it became the focus of discussion. This is his hands-on account plus community feedback.
On November 21, tech practitioner elvis posted on X: he is seriously considering switching his main everyday AI model from Claude Opus 5.5 to Gemini Neon. It's Google's newly released model, and his verdict was "Google absolutely cooked with this model" — they nailed it. After trying it across a combination of Pi and the Gemini API, he said this is the first Gemini model that felt to him like it could be a daily driver.
He also shared a visual explainer video: generated in a single pass with Gemini Neon, it covers the discussion of RSI in the first episode of a16z's podcast Working Theory, with voiceover from the ElevenLabs API. In his view, this kind of interactive visual explainer is becoming a great way to learn and digest research content.
He also posted a screenshot, saying the first thing he noticed was the output quality — interactions and notes were better than other models.
elvis has long preferred Gemini Flash for research output, notes, and explainers. The arrival of the new Neon made him want to move toward this new batch of models even more. He later posted two more visual explainer videos generated with Gemini Neon: the Riemann Hypothesis, and P vs NP.
In the comments, plenty of people agreed. One user said they already use it as a "baby AGI" and that it's quite reliable; another said Gemini has pulled off a crazy comeback over the past few months; someone else mentioned the Antigravity CLI experience and its generous limits. But it isn't without problems. In a reply, @nsxdavid posted a screenshot: when it comes to debugging code bugs, it performs poorly and isn't fast enough.
elvis also had observations about Google's evaluation strategy. benchmaxxing (optimizing to game benchmarks) isn't a good thing in itself, but this time Google may have found the signals that actually matter to users and optimized along those signals. That's how evals should be used. He included a screenshot.
For now he hasn't fully switched over. The sticking point is that Google's subscription can only be used within its own Antigravity, not across other toolchains. That limitation led elvis to say outright in a reply: it's time to launch a subscription that works across environments. If that opens up later, his switch will be very easy.
With this wave of new models, Google's release cadence going forward is worth watching. At least for one tester, Claude is no longer the default option.
发布时间: 2026-11-22 05:53