Artificial Analysis (cited): DeepSeek Flash model costs dramatically less to run
TNW ● Covered by 2 sources
DeepSeek’s V4-Flash now costs about 3 cents to run a full benchmark suite. That’s far below rival models, and it sharpens the AI price war.
Based on reporting by TNW — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
DeepSeek has a new price floor for frontier AI. According to Artificial Analysis, a version of its flagship model, V4-Flash, costs about three cents to run through the firm’s Intelligence Index benchmark suite. That makes it the cheapest well-known model the research group tested, and not by a little. Moonshot’s Kimi K3 came in at 86 cents, OpenAI’s GPT-5.6 Sol at $1.86, and Anthropic’s Claude Fable 5 at $3.15.
The published API pricing helps explain how DeepSeek gets there. The company charges $0.14 per million input tokens and $0.28 per million output tokens for the model. Flash is the cheaper half of DeepSeek’s pair of new releases, alongside the heavier V4-Pro, which is meant for harder reasoning. V4-Flash is the volume model: fast, cheap, and aimed at the kind of work that shows up everywhere, from chatbots to coding tools to back-office automation.
The tradeoff is capability, and DeepSeek does not hide it. V4-Flash scored 50 out of 100 on the Intelligence Index. That puts it level with Google’s Gemini 3.6 Flash and just behind Meta’s Muse Spark 1.1 and Z.ai’s GLM-5.2, which both scored 51. The top end is still ahead. Kimi K3 scored 57, while Claude Opus 5, Claude Fable 5, and GPT-5.6 landed roughly nine points higher again.
This lands in the middle of a price war DeepSeek has helped kick off. The company made a 75% discount permanent earlier this year, and rivals have been cutting too. OpenAI trimmed GPT-5.6 pricing sharply. The broader market is drifting down fast, in a way that looks less like normal software economics and more like a commodity fight.
DeepSeek has room to keep pushing. The company recently closed its first outside funding, a round of more than $7bn, which gives it space to subsidise aggressive pricing while it tries to win share. And the argument spreading through the market is simple: if a model that cheap can do most of the job, the expensive one has to justify those extra benchmark points in very plain English.
My take — AI-written commentary, not fact-checked reporting
This is what happens when AI turns into a utility: the bragging rights get cheaper, fast. The big Western labs can keep talking about premium quality, but the bill is starting to do more of the talking for them. Everyone loves a frontier model until the invoice shows up.
Read more about this at: TNW
