DeepSeek V4 Flash: open-weight MoE hits Intelligence Index 50 at $0.14/1M in

DeepSeek’s newly released V4 Flash 0731 lands at the top tier of open-weight reasoning models, scoring 50 on Artificial Analysis’s Intelligence Index — double the median of 25 for comparable models. It’s a Mixture-of-Experts design with 284 billion total parameters but only 13 billion active per token, shipped under a permissive MIT license with a 1M-token context window. The model is text-only, with no image input or multimodal support, and relies on extended chain-of-thought reasoning.

The headline story is cost. At $0.14 per million input tokens and $0.28 per million output tokens, it undercuts typical peers by a wide margin (medians around $0.43 and $1.20 respectively), and a full Intelligence Index evaluation ran to just $72. The main tradeoff is verbosity: the model burned 210 million tokens across the benchmark suite versus a 100M median, so its low per-token price is partly offset by how much it generates to reach an answer.

For teams weighing open-weight options, the combination of downloadable weights, commercial-friendly licensing, and frontier-adjacent reasoning scores makes it a strong candidate for self-hosting or cheap API use. Currently it’s served through a single API provider, and the elevated token consumption is worth budgeting for in reasoning-heavy or agentic workloads.