Meta's Muse Spark 1.1 undercuts OpenAI and Anthropic in AI price war

Meta just dropped Muse Spark 1.1, a multimodal reasoning model built for agent-based tasks like coding, computer use, and orchestrating multi-agent systems. The model handles a massive one-million-token context window and works in ‘Thinking’ mode within Meta’s apps. The big news? Meta is launching its own developer API for the first time, pricing output tokens at $4.25 per million — cheaper than OpenAI, Anthropic, and even xAI’s recently released Grok 4.5. This aggressive move pits Meta directly against pure-play AI labs that rely on high margins, squeezing them between well-funded tech giants and low-cost Chinese models. Muse Spark 1.1 also nails multi-agent orchestration, acting as a main agent that delegates tasks to subagents, improving coding and bug-fixing on large codebases. In benchmarks, it ranks fourth overall for speed and cost-efficiency, and jumped 36 spots on the coding benchmark. Why it matters: By undercutting competitors on price while offering enterprise-grade capabilities, Meta is reshaping the AI market — forcing rivals to rethink pricing and strategy as the war for developer wallets heats up.