Perplexity: Sonar

Perplexity Sonar is a text-and-image input model from Perplexity with a context window of roughly 127,000 tokens. It does not support tool use, reasoning modes, or structured output, which means it is suited to straightforward generation tasks rather than agentic pipelines or applications requiring schema-constrained responses. There is no published maximum completion length to plan around, so output size behavior is worth testing for your specific use case. At $1.00 per million tokens on both input and output, pricing is modest, but benchmark coverage is thin: a blended score of 14.3 drawn from only one benchmark leaves its general capability largely unproven by independent testing. Buyers who want a low-cost model for basic text or image-to-text tasks may find it worth a trial, but those making a high-stakes model selection should treat the benchmark standing as insufficient evidence and run their own task-specific evaluations before committing.