DeepSeek releases V4.1-Flash-1017 update over weekend, reviving AI efficiency fears and hitting Asian chip stocks

What happened

Chinese AI startup DeepSeek released DeepSeek-V4.1-Flash-1017 on Saturday, October 17, an update to the V4.1-Flash model launched September 10. DeepSeek calls it a Pareto improvement: faster task completion, shorter outputs, and better benchmark scores, with open weights published on Hugging Face under an MIT license. On its own benchmarks it scored 77.0 on the DeepSWE v1.1 coding test, up from 74.2 for the prior version and ahead of Anthropic's Claude Opus 5 (74.0) and OpenAI's GPT-5.6 Sol (73.0), at an effective cost under $0.20 per task versus $2 to more than $10 for U.S. rivals. DeepSeek also opened its in-house search engine to outside developers via Search and Fetch APIs. None of the benchmark figures have been independently verified, and Hugging Face co-founder Thomas Wolf publicly questioned whether the scores will hold up outside DeepSeek's own evaluation environment.

Who’s affected

• NVDA (Nvidia) — efficiency narrative directly questions demand for its accelerators; the January DeepSeek shock hit NVDA hardest
• AMD, AVGO, MU, INTC, AMAT — broad AI/chip complex under the same demand question; SOXX coming off a nine-day losing streak
• TSM (TSMC), SK Hynix, Advantest — dragged the Bloomberg Asia Semiconductor Index down 2%+ Monday to its lowest since July
• ANTHROPIC/OpenAI (private) — DeepSeek's own benchmarks claim to top Claude Opus 5 and GPT-5.6 Sol on coding/agent tasks at a fraction of per-task cost
• ANTHROPIC-adjacent: GOOGL, AMZN — investors in Anthropic; Anthropic IPO valuation chatter (~$1.5T vs $2T) already in play

Sources

siliconangle.comkucoin.comunbiasedheadlines.comxtb.com