DeepSeek Flash Matches GPT-5.6 Luna at 40% of the Price

DeepSeek has released the V4 Flash '0731' update: the model scores 50 on the Artificial Analysis Intelligence Index, compared to 51 for OpenAI's GPT-5.6 Luna. Each task costs 60% less — 284B parameters with only 13B active, a 1M-token context window, and an MIT license.

According to GDPval, the model climbed from 1,189 to 1,559 Elo. DeepSeek also offers a 98% discount on cached tokens versus the industry-standard 90% — making it especially attractive for repetitive agentic workloads.

The gap between Chinese budget models and American flagships has narrowed to a single benchmark point, even as the price difference remains 2x. That puts growing pressure on the entire AI market: companies paying for expensive APIs are finding it harder and harder to justify the cost.