Ling 3.0 Flash cuts AI costs without sacrificing reliability

Open-source model achieves lower per-task pricing while reducing hallucination rates

Ling 3.0 Flash, released by Ant Group's inclusionAI under MIT license, is the highest-scoring open model under 124 billion parameters on the Artificial Analysis Intelligence Index. It scores 38 points—outperforming its predecessor and matching Qwen3.6 27B in intelligence while using fewer active parameters. The model reduces hallucination from 97% to 44% on the AA Omniscience test and costs less per task than comparable models like Qwen3.6 27B. It also shows improved performance on banking tasks and achieves lower per-token pricing than similarly capable alternatives.

This model’s cost efficiency could lower barriers to accessible AI tools for individuals and small businesses. However, the metrics reflect specific benchmarks from August 13, 2026, and real-world impact remains unproven. The source notes that performance varies by task complexity—Ling 3.0 Flash consumes more tokens on complex tasks than alternatives—so practical savings depend on usage patterns. The model’s open availability via Hugging Face and DeepInfra supports wider adoption, but scaling effects on cost structures require further testing.

Source: The Decoder