Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-10-18
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
DeepSeek releases V4.1-Flash-1017, achieving stronger benchmarks, faster speed, and shorter outputs through a "Pareto improvement in performance and cost," which evaluators say creates competitive pressure on GPT 5.6 Sol [15][16][19]; meanwhile, Anthropic's Claude cracked the 25-year-old NSA ECDSA seed mystery in under 30 minutes, reigniting discussion about cryptographic backdoors and the boundaries of model reasoning [4].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Core Insights
**DeepSeek** has released **V4.1-Flash-1017**, achieving stronger benchmarks, faster speed, and shorter outputs through a "Pareto improvement in performance and cost," which evaluators say creates competitive pressure on **GPT 5.6 Sol** [15][16][19]; meanwhile, **Anthropic**'s **Claude** cracked the 25-year-old **NSA ECDSA seed mystery** in under 30 minutes, reigniting discussion about cryptographic backdoors and the boundaries of model reasoning [4].
## 🚀 Key Updates
- **DeepSeek releases V4.1-Flash-1017** [19]: Stronger benchmarks, faster tasks, shorter outputs; now available on the official website, App, and API, with its internal search engine opened as the Search & Fetch API.
- **Claude cracks 25-year-old cryptography puzzle** [4]: Anthropic's Claude solved the mystery of the ECDSA seed origin—standardized by NIST in 2000 and provided by the NSA—in under 30 minutes.
- **Arena Alignment Index adds three new models** [1]: Meta Muse Wonder (Max) ranks 6th at 77.5%, with Reflection Beam-1.0-501B-A23B and StepFun Step 5 also joining the list.
- **Meta removes controversial scraping clause** [2]: After experts interpreted its terms as allowing scraping of news for AI training and overriding robots.txt, Meta updated its commercial terms to remove the relevant wording.
- **Trend of websites blocking AI agents intensifies** [13]: Yelp only allows paid licensed agent access; Delta, JPMorgan Chase, and Charles Schwab block access on security grounds; Cars.com and Amtrak are already unavailable.
- **Glimpse visual reasoning model released** [7]: The Elorian team claims its visual reasoning rivals Project Astra, with API costs reduced by 17x, tackling spatiotemporal physical causal reasoning.
- **Multiple domestic AI teams have not released new models for over 3 months** [11]: ByteDance's Doubao-Seed-2.1-Pro, MiniMax-M3, and Moonshot's Kimi-K3 have all gone more than 3 months without updates.
- **Agentic Commerce may replace 20–30% of commerce** [0]: Analysis suggests it will replace 20%–50% of existing commerce while expanding the overall pie by about 50%.
## 🔗 Sources
[0] Agentic Commerce may replace 20–30% of commerce — https://aihot.news/items/col0apvosdtys8kqjuwx68wk5
[1] Arena Alignment Index update: Meta, Reflection, and StepFun three new models on the list — https://aihot.news/items/v27j8fnpclhpwv5blr21n4650
[2] Meta removes wording that could be interpreted as allowing scraping of news websites for AI training — https://aihot.news/items/ib3axwzp3wkqdd8teu8zpj7c9
[4] Anthropic's Claude cracks the NSA's 25-year-old ECDSA seed mystery in under 30 minutes — https://aihot.news/items/gyb0pby5x9ou5m8szgpr0k4vh
[7] Glimpse: A low-cost, high-capability visual reasoning model — https://www.bestblogs.dev/status/2111454725577032125?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[11] Multiple domestic AI teams have not
DeepSeek has released V4.1-Flash-1017, achieving stronger benchmarks, faster speed, and shorter outputs through a "Pareto improvement in performance and cost," which evaluators say creates competitive pressure on GPT 5.6 Sol [15][16][19]; meanwhile, Anthropic's Claude cracked the 25-year-old NSA ECDSA seed mystery in under 30 minutes, reigniting discussion about cryptographic backdoors and the boundaries of model reasoning [4].
🚀 Key Updates
- DeepSeek releases V4.1-Flash-1017 [19]: Stronger benchmarks, faster tasks, shorter outputs; now available on the official website, App, and API, with its internal search engine opened as the Search & Fetch API.
- Claude cracks 25-year-old cryptography puzzle [4]: Anthropic's Claude solved the mystery of the ECDSA seed origin—standardized by NIST in 2000 and provided by the NSA—in under 30 minutes.
- Arena Alignment Index adds three new models [1]: Meta Muse Wonder (Max) ranks 6th at 77.5%, with Reflection Beam-1.0-501B-A23B and StepFun Step 5 also joining the list.
- Meta removes controversial scraping clause [2]: After experts interpreted its terms as allowing scraping of news for AI training and overriding robots.txt, Meta updated its commercial terms to remove the relevant wording.
- Trend of websites blocking AI agents intensifies [13]: Yelp only allows paid licensed agent access; Delta, JPMorgan Chase, and Charles Schwab block access on security grounds; Cars.com and Amtrak are already unavailable.
- Glimpse visual reasoning model released [7]: The Elorian team claims its visual reasoning rivals Project Astra, with API costs reduced by 17x, tackling spatiotemporal physical causal reasoning.
- Multiple domestic AI teams have not released new models for over 3 months [11]: ByteDance's Doubao-Seed-2.1-Pro, MiniMax-M3, and Moonshot's Kimi-K3 have all gone more than 3 months without updates.
- Agentic Commerce may replace 20–30% of commerce [0]: Analysis suggests it will replace 20%–50% of existing commerce while expanding the overall pie by about 50%.
🔗 Sources
[0] Agentic Commerce may replace 20–30% of commerce — https://aihot.news/items/col0apvosdtys8kqjuwx68wk5
[1] Arena Alignment Index update: Meta, Reflection, and StepFun three new models on the list — https://aihot.news/items/v27j8fnpclhpwv5blr21n4650
[2] Meta removes wording that could be interpreted as allowing scraping of news websites for AI training — https://aihot.news/items/ib3axwzp3wkqdd8teu8zpj7c9
[4] Anthropic's Claude cracks the NSA's 25-year-old ECDSA seed mystery in under 30 minutes — https://aihot.news/items/gyb0pby5x9ou5m8szgpr0k4vh
[7] Glimpse: A low-cost, high-capability visual reasoning model — https://www.bestblogs.dev/status/2111454725577032125?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[11] Multiple domestic AI teams have not
← Back to Updates