Stock Markets August 3, 2026 01:42 AM

DeepSeek’s V4-Flash undercuts rivals on running costs while matching mid-tier benchmark scores

Research firm finds the Chinese startup’s new model far cheaper per test than major peers, even as competitors and larger models outperform on intelligence benchmarks

By Nina Shah
Share
Twitter Reddit Facebook LinkedIn
GOOGL META BABA

Artificial Analysis, a San Francisco research firm, found DeepSeek’s newly released V4-Flash model to be substantially less expensive to run on benchmark tests than widely known AI models, with an estimated average cost of $0.03 per test. V4-Flash charges $0.14 per million input tokens and $0.28 per million output tokens, and scored 50 on an Intelligence Index of nine benchmarks, comparable with Google’s Gemini 3.6 Flash. The finding highlights a price-versus-performance trade-off as DeepSeek seeks to regain momentum amid intense domestic competition and prepares a more powerful V4-Pro.

DeepSeek’s V4-Flash undercuts rivals on running costs while matching mid-tier benchmark scores
GOOGL META BABA
Summarize with
ChatGPT Perplexity Claude Grok Gemini

Key Points

  • DeepSeek’s V4-Flash has an estimated average running cost of $0.03 per test based on Artificial Analysis’s benchmarks, priced at $0.14 per million input tokens and $0.28 per million output tokens.
  • On the Intelligence Index of nine benchmarks, V4-Flash scored 50, equal to Google’s Gemini 3.6 Flash and below several higher-scoring models including Moonshot’s Kimi K3 and offerings from Anthropic and OpenAI.
  • The model launches amid fierce competition from Chinese startups and large tech firms as companies and enterprises look to balance lower deployment costs against model capability.

Overview

DeepSeek has formally released V4-Flash, a new iteration of its flagship AI model aimed at offering an ultra-low-cost option for businesses deploying generative AI. According to a San Francisco research firm, Artificial Analysis, V4-Flash is materially cheaper to run on benchmark tests than a range of well-known models worldwide.

Pricing and measured running cost

Artificial Analysis reported that V4-Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens. The research group calculated an average running cost of roughly $0.03 per test for V4-Flash on its benchmark suite. That figure compares with $0.86 per test for Moonshot AI’s Kimi K3, $1.86 for OpenAI’s GPT-5.6 Sol, and $3.15 for Anthropic’s Claude Fable 5.

The research firm emphasized that this cost comparison factors in the data a model must process and generate to complete tasks, noting that headline price alone can be misleading if a model requires many more steps to produce an answer.

Performance on intelligence benchmarks

On Artificial Analysis’s Intelligence Index, which aggregates results from nine benchmarks covering coding, reasoning and workplace-style tasks, V4-Flash scored 50 out of 100. That score matches Google’s Gemini 3.6 Flash and sits one point behind Meta’s Muse Spark 1.1 and GLM-5.2 from Z.AI, also known as Zhipu.

Moonshot’s Kimi K3 scored 57 on the same index. Anthropic’s Claude Opus 5, Claude Fable 5, and OpenAI GPT-5.6 each scored nine or more points higher than V4-Flash on the Intelligence Index.

Market context and competitive landscape

DeepSeek rose to prominence with its earlier R1 model, which became a global headline in early 2025 and prompted a selloff in global technology stocks while raising questions about the scale of U.S. companies’ AI spending. Since then, the firm has faced intense domestic competition from startups such as Moonshot, MiniMax and Z.AI, along with larger Chinese technology companies including ByteDance and Alibaba. These players, together with major U.S. firms, are competing for enterprise customers seeking lower-cost ways to deploy AI at scale.

DeepSeek is reportedly preparing a more capable variant called V4-Pro, though it has not provided a release date for that model. Separately, Alibaba on Monday unveiled its largest and most capable AI model to date, Qwen3.8-Max, which the article notes is not far behind in size compared with a Moonshot AI offering launched last month.

Implications for customers and vendors

The Artificial Analysis comparison aims to provide a practical measure of value by combining cost and the data processing required to complete tasks. For businesses evaluating AI deployments, the findings suggest that headline pricing should be weighed against per-test efficiency and benchmarked task performance. V4-Flash’s low running cost may appeal to organizations focused on economics of scale, while other models retain advantages on higher benchmark scores.


Key points

  • DeepSeek’s V4-Flash is reported to be the least expensive to run among well-known models in the Artificial Analysis benchmark set, at an estimated $0.03 per test.
  • V4-Flash scored 50 on the Intelligence Index, matching Google’s Gemini 3.6 Flash and trailing several higher-scoring models; Moonshot’s Kimi K3 scored 57, while Anthropic and OpenAI models scored notably higher.
  • The model is entering a crowded market of Chinese startups and large tech firms competing for enterprise adoption and lower-cost AI deployment.

Risks and uncertainties

  • Benchmark scores indicate that some competing models outperform V4-Flash on intelligence tasks, creating a trade-off between cost and capability that could affect adoption decisions in enterprise IT and cloud services.
  • No release date has been given for the more powerful V4-Pro, leaving uncertainty about DeepSeek’s near-term product roadmap and competitive positioning.
  • Intense domestic competition from startups and major Chinese tech companies may pressure DeepSeek’s market share and fundraising or IPO plans in the absence of clearly superior performance.

Notes: All pricing, benchmark scores and comparative cost-per-test figures are those published by Artificial Analysis and as reported in the company statements noted above. Token prices refer to units of data used to measure AI usage.

Risks

  • Higher-scoring models may be preferred for capability-sensitive applications, potentially limiting V4-Flash adoption in enterprise IT and cloud services.
  • Timing for the promised V4-Pro is unspecified, creating uncertainty about DeepSeek’s near-term product competitiveness and roadmap.
  • Domestic competition from other Chinese startups and major tech firms could pressure DeepSeek’s market position and any potential IPO plans.

More from Stock Markets

Taiwan market ends flat as select stocks swing sharply at close Aug 4, 2026 Apple Seeks Emergency Court Order to Block OpenAI and Ex-Employees from Alleged Trade Secret Use Aug 4, 2026 Wage Talks Stall at Port Hedland as Unions Advance Weekend Strike Plan Aug 4, 2026 Australian lithium shares jump after Liontown presentation brightens sector outlook Aug 4, 2026 Rio Tinto not expected to reopen Glencore merger talks as standstill lapses Aug 4, 2026