On September 2, Meta and Google released competing AI models within hours of each other. Meta's Muse Spark 1.3 scores 62 on the Intelligence Index, trailing only Claude's top models, at $1.25 per million input tokens. Google's Gemini 3.8 Flash scores 59, costs $0.75 per million tokens through December, then doubles in January.
The efficiency war
Meta says Spark 1.3 uses 20% fewer tool calls and 25% fewer tokens than its predecessor. Google says 3.8 Flash works harder on difficult tasks, using extra reasoning steps. Independent testing found Google's model costs roughly 40% more per completed task because it burns more tokens, even though the sticker price stayed flat.
Meta's pricing trick
Meta offers a contributor tier that costs 10 to 20 times less than standard pricing. The trade: Meta trains on your data. A meaningful double-digit percentage of developers already chose that option, according to Meta's AI lead.
Both models target the same developers: teams building coding agents and long-running automated workflows who need something cheaper than Claude but smarter than last year's models. The sticker price looks like a deal. The real cost depends on how many tokens the job burns and whether you are willing to let the vendor learn from your work.