AI RECAP: Rumored GPT-4o Large Model & Gemini Live vs GPT-4o Advanced Voice

A swirl of “strawberry/Q*” rumors about OpenAI’s next reasoning model is colliding with concrete updates—yet the most important question remains unanswered: whether any new model actually changes what users can do inside ChatGPT. The speculation centers on a rumored “Q* / strawberry” architecture described as a new reasoning engine meant to give large language models more human-like problem solving. A simple test is used to illustrate the gap today: asking a model how many “R” letters are in “strawberry” often produces an incorrect count, even though it can sometimes self-correct after being prompted.

The hype takes on a more concrete feel because high-profile OpenAI figures appear to engage with strawberry-themed posts on X. Sam Altman replies “amazing” to a tweet from an account using strawberries as its branding, and other AI insiders amplify the attention. At the same time, the official ChatGPT account posts about a “new GPT-4o Omni model” being available in ChatGPT, but the transcript’s author reports not noticing any obvious differences in the app—no clear model switch, no access to new “Advanced voice,” and no change in image generation behavior (still tied to the Dolly3 API and existing image tools).

Still, one prediction seems to land on timing. The strawberry-themed account had claimed something would arrive from OpenAI at 10:00 a.m., and OpenAI did release an updated SWE-bench iteration—a benchmark designed to evaluate how well AI systems solve real-world software tasks. That matters because it’s a more developer-relevant yardstick than generic chat benchmarks. The same account also points to Thursday as a likely day for a “GPT-4o Omni large” release, with the idea that OpenAI might pair a new model with a benchmark drop to justify performance gains.

Even if a Thursday release arrives, the transcript argues that “better reasoning” won’t be convincing without a clear explanation of how it works and when it beats alternatives like prompting tricks or retrieval/search-based systems such as Perplexity. The core frustration is transparency: users want to know what’s actually changing under the hood, not just what’s being teased.

On the Google side, the day’s major announcement is “Gemini Live,” positioned as a voice-first experience on Android. The transcript’s author demonstrates a Gemini Live flow with multiple selectable voices and a conversation that brainstorms science activities, but then draws a key distinction: Gemini Live appears to convert speech to text, generate a response in text, and then convert that text back to audio—meaning it’s not “native” multimodal voice in the same way as OpenAI’s rumored/advanced voice approach. The Gemini Live feature set also includes Android integration and small utilities (like a calendar extension that reads concert flyers), but nothing presented as a leap that would force a platform switch. Overall, the transcript frames OpenAI’s strawberry/Q* mystery as the more consequential story, while Google’s Gemini Live looks incremental compared with what’s already available elsewhere.