DeepSeek's V4-Flash is cheapest well-known AI model to run, study finds

DeepSeek's newly released V4-Flash model is by far the cheapest well-known AI model to run, according to research firm Artificial Analysis, which tested it against rivals from OpenAI, Anthropic and others.

Artificial Analysis said the San Francisco firm found V4-Flash averaged 3 cents per test, compared with 86 cents for Moonshot's Kimi K3, $1.86 for OpenAI's GPT-5.6 Sol and $3.15 for Anthropic's Claude Fable 5, making it more than 100 times cheaper to run than Fable 5.

DeepSeek released V4-Flash on Friday 31 July 2026, and Reuters reported the cost findings on 3 August, according to the report via Yahoo Finance.

The model's list price is $0.14 per million input tokens and $0.28 per million output tokens, Artificial Analysis said.

Artificial Analysis said it measures cost per test rather than list price alone, because a model that looks cheap per token can still cost more overall if it needs many more steps to reach a correct answer.

On performance, V4-Flash scored 50 out of 100 on the Artificial Analysis Intelligence Index, a measure across nine benchmarks covering coding, reasoning and workplace tasks, according to The Next Web. That places it level with Google's Gemini 3.6 Flash, one point behind Meta's Muse Spark 1.1 and Z.ai's GLM-5.2, and at least nine points behind Claude Opus 5, Fable 5 and GPT-5.6. Kimi K3 scored 57.

V4-Flash uses a mixture-of-experts architecture with about 284 billion total parameters and about 13 billion active per request, hybrid attention, and a 1 million token context window, Artificial Analysis said.

DeepSeek is reportedly preparing for a possible initial public offering. Its R1 model triggered a global tech sell-off in early 2025. Analysts have not yet said whether V4-Flash's pricing will affect competitors' strategies.