The AI launch calendar got crowded. On Sep 3, 2026, Meta and Google joined the AI launch party with new models, giving the market two fresh claims about performance, price, and practical use. Meta introduced Muse Spark 1.3, while Google launched Gemini 3.8 Flash on Wednesday.
Meta is pitching Spark 1.3 as “frontier performance almost too cheap to meter.” The model’s Max version scored 62 on AA’s Intelligence Index, putting it three points ahead of Gemini 3.8 Flash, which scored 59.
Tulsee Doshi said Spark 1.3 models “really surprised us in positive ways in their performance.” She also described Meta’s aim: “We’re really excited about being able to provide an offering to defenders that is a fraction of the cost, much faster, while still showcasing that frontier-level performance.”
Meta’s roadmap already has another model waiting behind the curtain. Mark Zuckerberg said “next up” is Meta’s larger model codenamed “Watermelon,” because apparently naming advanced systems after fruit remains a boardroom-approved strategy.
Google’s Gemini 3.8 Flash takes a different route, focusing on coding and agentic tasks while keeping Gemini 3.7’s pricing. It costs 75 cents per million input tokens and $3.75 per million output tokens — listed as $0.75 / $3.75.
That pricing gives Google a clear point of continuity, but the model’s positioning is more complicated. DeepMind executive Koray Kavukcuoglu admitted Gemini sits “a little below the frontier,” then added, “there’s nothing other than being at the frontier that’s important for us.”
Gemini 3.8 Flash still represents a stronger bounce-back from Google’s 2026 struggles. The company’s stock rose 0.6% on Wednesday after dropping more than 1% Tuesday, against a four-month losing streak that had made every market move look like a referendum on its AI strategy.
Google’s AI case also rests on usage inside its cloud business, not only on model scores. Nearly three-quarters of Google Cloud customers are already using its AI products, and Google Cloud CEO Thomas Kurian said those customers are spending about 50% more than their original commitments.
That customer spending gives Gemini a commercial base while Google works through questions about model leadership. It also offers a response to the familiar AI problem: impressive demonstrations matter less when businesses do not keep paying for them.
Google Gets Breathing Room Beyond Models
Google received a separate boost on Wednesday when a federal judge rejected the Justice Department’s push to force the company to sell its ad exchange. The decision leaves Google’s ad business intact as its latest-quarter revenue grew 14%.
The legal outcome does not erase the antitrust fight, but it changes the immediate pressure on Google’s advertising operation. Combined with the stock move and cloud spending, it gives the company a better business backdrop while Gemini tries to close the gap with frontier models.
The broader AI story now looks less like a single-model race and more like a contest across several scoreboards. Meta has Spark 1.3’s 62 on AA’s Intelligence Index and a larger Watermelon model “next up,” while Google has a 59-scoring Flash model aimed at coding, agents, and cloud customers willing to spend.
That split matters because low prices alone do not make a model useful, and benchmark scores alone do not make it a business. Meta is selling cost and performance together; Google is tying Gemini to its cloud products, its advertising engine, and a company that has just gained room to keep playing.
The launch wave also drew 3,428 new readers, another reminder that AI releases remain attention magnets even when the technical rankings refuse to produce a clean winner. For now, September has delivered two new models, one legal reprieve, and plenty of evidence that the frontier is still a moving target.
Based on
