Google's Gemini 4 Isn't Out Yet — How to Spot the Fake Rumors

Here's what Google has actually confirmed: it began training Gemini 4 on 21 July 2026, calling it its "most ambitious" training run yet. CEO Sundar Pichai added that it will be built on a much larger base model, with coding and AI agents (software that takes actions for you) as priorities. That's it. No date, no price, no specs. And "training has started" is not the same as "almost ready" — safety testing and fine-tuning come afterward, which can take months.

Every other detail you've seen online came from social media. The "October launch," the "argon" codename, the claim it can handle 2 million tokens (chunks of text) — all trace back to anonymous X accounts with no Google source. Here's a simple filter: if the number isn't on blog.google or ai.google.dev, it's a guess. Some sites are publishing tidy Gemini 4 pricing tables for a model that doesn't exist, purely to grab search traffic.

So what can you use today? Gemini 3.8 Flash, released 2 September 2026. It reads about 1 million tokens at once — roughly 750,000 words, longer than most novels — and costs $0.75 per million tokens of input and $3.75 for output. That promotional pricing ends 31 December 2026, after which it doubles to $1.50 and $7.50. If you're budgeting for AI tools, plan for that increase.

When will Gemini 4 arrive? Google's recent rhythm offers a clue: Gemini 3 Pro in November 2025, Flash in December, 3.5 Flash-Lite in July 2026, 3.8 Flash in September. Pichai says releases are becoming nearly monthly. So a late-2026 flagship is plausible — but unconfirmed. Watch three things on launch day: whether it reads more than 1 million tokens, whether it writes longer than 64,000 tokens at a stretch, and the price.