Meta has officially overtaken OpenAI in head-to-head reasoning and latency benchmarks with the release of Muse Spark, a multimodal model that hit a 92.4 MMLU-Pro score—edging out GPT-5.2's 91.8. The model is 2.3x cheaper per token than OpenAI's comparable tier, and early enterprise deployments show 38% faster inference on standard cloud infrastructure.
- Core Update: Meta Muse Spark released globally, surpassing OpenAI on reasoning benchmarks with a 92.4 MMLU-Pro score.
- Key Metrics / Specs: 2.3x cheaper per token than OpenAI's flagship, 38% faster inference, and native Hindi + Tamil + Telugu support out of the box.
- Access & Availability: Live today via Meta's API and Azure Marketplace; pricing starts at $0.55 per million input tokens—no waitlist for Indian regions.
For Indian startups and SMBs, the immediate win is cost-per-call drops of up to 60% for customer service and document-processing agents. Muse Spark's native Indic language fluency means chatbots no longer need a separate translation layer—a major advantage for Chennai-based firms managing multilingual client bases. The model also handles complex tool-calling sequences with a 96% tool-selection accuracy, making multi-step automation workflows far more reliable.
This shift signals a broader trend: enterprises are no longer locked into a single AI provider. Teams building on OpenAI can now port prompts and fine-tuned weights to Muse Spark with minimal refactoring, giving Indian businesses leverage in contract negotiations. For agencies and internal tech teams exploring options, this also validates a multi-model strategy—especially for regional SEO and hyper-local content generation where Meta's Indic-language training data excels. Businesses looking to capitalize on this shift can benefit from AI Agents & Bots tailored to these new models.