DeepSeek launches V4.1 Flash, will retire V4 Pro tier

Chinese AI startup DeepSeek on Thursday released DeepSeek-V4.1-Flash, a 552-billion-parameter model it described as the smallest in a new architecture family, pairing the launch with lower API prices and a plan to phase out its more expensive V4 Pro tier.reuters+1

The model uses what DeepSeek calls an asymmetric Causal Encoder-Decoder structure built on a mixture-of-experts design, with only 8 billion parameters active for processing input and 16 billion for generating output — a split intended to keep inference costs low. It ships with native multimodal visual understanding, folding in image support that previously required a separate experimental model.api-docs.deepseek+1

Benchmarks and Pricing

DeepSeek's own benchmark table shows V4.1 Flash surpassing its V4 Pro predecessor and trading blows with OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5 on several agentic and coding tasks. On DeepSWE v1.1, a software engineering benchmark, V4.1 Flash scored 74.2, narrowly ahead of Claude Opus 5 at 74.0 and GPT-5.6 Sol at 73.0. On knowledge-heavy evaluations like HLE, however, Claude Opus 5 still holds a clear lead. Independent benchmark boards have not yet published scores for the new model.intelligentliving+1

The new pricing, effective September 10 at noon Beijing time, sets off-peak output at $0.60 per million tokens and uncached input at $0.15 per million tokens. That represents a roughly 9% drop in output cost and a 32% drop in uncached input cost from the post-hike V4 Flash rates DeepSeek introduced on August 16. The August price increase — which raised some rates by more than 10 times — drew sharp criticism from developers and came under additional pressure from open-source clones that undercut DeepSeek's official API.qz+4

Pro Routing and IPO Preparations

Starting September 14 at noon Beijing time, all API requests directed at V4 Pro will be automatically routed to V4.1 Flash and billed at Flash-series rates, effectively retiring the Pro tier until a future V4.1 Pro model arrives. DeepSeek said internal and external testing confirmed V4.1 Flash had surpassed V4 Pro on performance, cost, speed, and total completion time.news.aibase+1

The launch comes as DeepSeek prepares for an initial public offering on Shanghai's STAR Market. The company has hired CITIC Securities to underwrite the listing and aims to begin the IPO process this year, according to Reuters, though the timing, size, and target valuation remain undecided. A recent pre-IPO financing round valued the Hangzhou-based company at roughly $74 to $75 billion.datastudios+2