DeepSeek unveils v4.1 Flash, offering more power at a lower price than v4 Pro
deepseek
| Source: HN | Original article
DeepSeek announced the launch of its V4.1 Flash model, which is cheaper and more capable than the V4 Pro, with new pricing taking effect at noon Beijing time on Sept 10, 2026.
DeepSeek, the Chinese AI lab behind the popular “Flash” family of language models, has rolled out v4.1 Flash, a version it claims is both cheaper and more capable than its flagship V4‑Pro. The company announced a pricing adjustment effective 12:00 Beijing time on 10 September 2026: during off‑peak hours the unit price will be $0.003 for input cache hits, $0.15 for input cache misses and $0.60 for output tokens, with peak‑hour rates simply doubled. Earlier reports placed Flash’s cost at roughly $0.14 per million input tokens and $0.28 per million output tokens – about a third of V4‑Pro’s price – and the new schedule reinforces that advantage.
Beyond price, v4.1 Flash adds native multimodal support and a refreshed architecture that delivers faster processing and higher benchmark scores on coding and agentic tasks, according to early testers. The internal test version, released on a two‑day limited beta, retains the original base URL, allows up to 20 concurrent requests per account, and is priced identically to the existing “deepseek‑v4‑flash” endpoint. DeepSeek’s messaging suggests the model is intended to absorb workloads that previously required the more expensive V4‑Pro, potentially reshaping how developers allocate compute budgets.
The launch matters because it intensifies price competition in the generative‑AI market, where providers such as Anthropic and OpenAI have faced scrutiny over cost and accessibility. A cheaper, multimodal model could accelerate adoption in cost‑sensitive sectors across the Nordics, from fintech to media production, and may pressure rivals to revisit their own pricing structures.
What to watch next: how the limited beta performs under real‑world demand, whether DeepSeek extends the off‑peak pricing beyond the trial period, and how the model’s multimodal capabilities compare on public benchmarks. Industry observers will also be keen to see if the cost advantage translates into broader enterprise contracts, especially as European regulators keep a close eye on AI pricing and market dynamics.
Sources
Back to AIPULSEN