Gemini 4 Pro's first checkpoint was reported on September 4 by leaker lyra, who said Google had "finally" released it after "a long time in testing." A public launch is expected in October. Google CEO Sundar Pichai confirmed in July that Gemini 4 was in training but gave no timeline.
The checkpoint's existence does not mean the model is finished. It means Google is testing internally. The gap between checkpoint and launch is where models either become products or get canceled.
Google Canceled Its Last Flagship. This Is the Only Bet Left.
Google's flagship pipeline has been disrupted. Gemini 3.5 Pro, originally promised for June, was quietly canceled after internal candidates failed to show sufficient improvement over the Flash series. The company redirected its resources toward Gemini 4.
That decision created a months-long gap at the top of Google's model lineup. The Flash series has kept iterating—Gemini 3.6 Flash in July, 3.7 Flash in August, 3.8 Flash in September—but Flash models are workhorses, not flagships. They narrow the coding gap; they don't close it.
Gemini 4 Pro is now Google's only path to reclaiming the frontier. If it slips again, the gap widens.
Flash-Lite First. Flagship Later.
Before Gemini 4 Pro arrives, Google is shipping lighter models. Lyra expects a Gemini 4 Flash-Lite and an updated NB2Lite image model in September. The strategy is consistent with Google's recent cadence: iterate frequently on small models to maintain developer mindshare, and hold the flagship until it's ready.
This is a volume play. Flash-Lite models are cheaper and faster to deploy. They penetrate developer ecosystems where full-size models are too expensive or too slow. By the time Gemini 4 Pro launches, Google hopes those developers are already building on its platform.
The risk is that "good enough" becomes the ceiling. If developers are satisfied with Flash-Lite, the flagship's arrival matters less.
The Leaked Benchmarks Say "Predicted"—Not Measured
Leaked evaluations circulating on social media claim Gemini 4 Pro beats Fable 5.1 on autonomous multi-file refactoring and GPT-6 Astra on multi-step reasoning chains, with native support for up to 1.5 million tokens of context.
The fine print matters. The benchmark tables are labeled "PREDICTED." There is no checkpoint ID, no inference configuration, no test methodology. These are expectations, not results. Google has not published official benchmarks, and the leaker is an individual account, not a verified source.
If the predictions hold, Gemini 4 Pro could be Google DeepMind's most decisive release. If they don't, the wait was for nothing.
Google is using lightweight models to hold the line while its flagship catches up. The strategy buys time. It does not guarantee victory. Gemini 4 Pro's October launch will be judged against Fable 5.1, GPT-6 Astra, and Muse Spark 1.3—all shipping now. The competition isn't waiting for Google to finish training. The real race is not one flagship launch. It is whether Google can ship enough Flash-Lite models to keep developers on its platform until the flagship arrives.
P.S. The checkpoint leak says Gemini 4 Pro is "significantly more capable" than Gemini 3.5 Pro would have been. That is the entire justification for canceling 3.5. If the October launch is delayed again, the decision to abandon 3.5 will look less like a strategic pivot and more like a missed cycle.
Frequently Asked Questions
Q: What is Gemini 4 Pro's first checkpoint?
A: The first internal checkpoint of Gemini 4 Pro was reported on September 4 by leaker lyra. A checkpoint is an intermediate model version used for internal testing before public release. Google has not officially confirmed its existence.
Q: When will Gemini 4 Pro launch?
A: The leaker expects a public launch in October 2026. Google CEO Sundar Pichai confirmed in July that Gemini 4 was in training but gave no timeline.
Q: What happened to Gemini 3.5 Pro?
A: Gemini 3.5 Pro was canceled. Internal candidates failed to show sufficient improvement over the Flash series, and Google redirected its resources toward Gemini 4.
Q: What other models is Google shipping before Gemini 4 Pro?
A: Google is expected to release a Gemini 4 Flash-Lite and an updated NB2Lite image model in September 2026.
Q: What do the leaked benchmarks show?
A: Leaked evaluations claim Gemini 4 Pro beats Fable 5.1 on autonomous multi-file refactoring and GPT-6 Astra on multi-step reasoning chains, with native support for up to 1.5 million tokens of context. However, the benchmark tables are labeled "PREDICTED" and lack verification details.
Q: Are the leaked benchmarks reliable?
A: The benchmarks are labeled "predicted," not measured. There is no checkpoint ID, no inference configuration, and no test methodology. Google has not published official benchmarks. These are expectations, not verified results.
Q: Why did Google cancel Gemini 3.5 Pro?
A: Internal candidates for Gemini 3.5 Pro failed to show sufficient improvement over the Flash series. Google redirected its resources toward Gemini 4.
Q: What is Google's strategy with Flash-Lite models?
A: Google is iterating frequently on small models to maintain developer mindshare while holding the flagship until it's ready. Flash-Lite models are cheaper and faster to deploy, penetrating developer ecosystems where full-size models are too expensive.
Q: What is the risk of this strategy?
A: If developers are satisfied with Flash-Lite models, the flagship's arrival matters less. "Good enough" becomes the ceiling.
Q: What will Gemini 4 Pro compete against?
A: Gemini 4 Pro will be judged against Fable 5.1 (Anthropic), GPT-6 Astra (OpenAI), and Muse Spark 1.3 (Meta)—all shipping now. The competition isn't waiting for Google to finish training.
