DeepSeek, a Chinese AI startup, officially launched its flagship model V4-Flash on July 31, 2026, introducing a highly cost-efficient AI option to the market [1, 2, 3, 4, 5]. V4-Flash operates on a Mixture of Experts (MoE) architecture with 284 billion total parameters, of which 13 billion are activated at any time [3, 5].
Cost analysis by Artificial Analysis estimates the operating cost for V4-Flash at $0.14 per million input tokens and $0.28 per million output tokens. The average cost to complete a benchmark test is roughly $0.03 [1, 2, 6, 4, 5]. By comparison, Moonshot AI’s Kimi K3 costs about $0.86, OpenAI’s GPT-5.6 Sol costs $1.86, and Anthropic's Claude Fable 5 costs $3.15 per test, making V4-Flash more than 100 times cheaper than some competitors [1, 2, 6, 4, 5]. Artificial Analysis noted that "this comparison, which includes data volume processed and generated, better reflects the true usage cost of AI models" [2].
On the Intelligence Index, V4-Flash scored 50 out of 100 across nine benchmark tasks covering programming, reasoning, and workplace simulations [1, 2, 6, 4, 5]. This matches Google Gemini 3.6 Flash but trails behind Meta’s Muse Spark 1.1, Zhipu AI’s GLM-5.2, and significantly behind Kimi K3 and GPT-5.6, which scored at least 9 points higher [1, 2, 6, 4, 5]. Despite the lower score, DeepSeek officials stated, "We aim to leverage our low-cost advantage to regain momentum in the AI market" [1].
V4-Flash supports the OpenAI Response API format and is compatible with OpenAI Codex, allowing developers to switch models with minimal changes [3]. In response to DeepSeek’s launch, OpenAI announced an 80% price cut for its GPT-5.6 Luna model in late July 2026 [3, 5].
Following the release, usage data showed V4-Flash reaching 7.22 trillion tokens per week on the OpenRouter API platform and hitting 8 trillion tokens in a single day on the OpenCode platform, indicating strong adoption [5].
Reports suggest DeepSeek is preparing for a potential initial public offering in the wake of the V4-Flash launch [1, 2, 4].
