Sam's News β anthropic β 2026-10-08¶
AI¶
7.5 Claude Haiku 5.5 offers up to 90% token price reduction versus GPT-6 Luna¶
Anthropic's Claude Haiku 5.5 delivers substantial token cost cuts and claims performance advantages over OpenAI's GPT-6 Luna.
Sources: Intelligent Living RSS, WinBuzzer RSS, Adgully.com RSS, India Today RSS, Gadgets 360 RSS, Root-Nation.com RSS Update to: Claude Haiku 5.5 released
7.5 Anthropic unveils Claude Haiku 5.5 model, third release in a month ahead of planned IPO¶
Anthropic released Claude Haiku 5.5, its third model launch within a month as the company prepares for a public offering.
Sources: The Indian Express RSS
7 Anthropic cuts Claude pricing 75% ahead of planned $2 trillion IPO valuation¶
Anthropic cut Claude cache read pricing 75% to $0.25 per million tokens, matching OpenAI's base pricing as the company prepares for an IPO valued at up to $2 trillion. The company reported $4.6 billion 2025 revenue (12x 2024) but a $42 billion loss, with $518 billion in committed computing capacity.
- Cache read pricing cut 75%: $1 to $0.25 per million tokens
- Base pricing: $10 input, $50 output per million tokens (matches GPT-6 Astra)
- 2025 revenue: $4.6 billion; operating loss: >$8 billion
- IPO targeted at $2 trillion valuation, November 2026 listing
Sources: 24/7 Wall St. AI Web Searched, BigGo Finance AI Web Searched, BeInCrypto RSS
5 Anthropic Expands Access to Claude Mythos AI Model¶
Anthropic announced broader availability of its Claude Mythos language model.
Sources: NHKγγ₯γΌγΉ RSS
5 Anthropic expands Claude AI with Chinese-language options¶
Anthropic added Chinese-language configuration options to the Claude AI platform.
Sources: South China Morning Post RSS
4.5 Anthropic's Claude Max and Team Plans Include API Credits With Claude Code Restrictions¶
Anthropic adds API credits to its Claude Max and Team subscription plans, though these credits cannot be used with the Claude Code feature.
Sources: MIXED Reality News RSS
Security¶
7.5 Chinese Hacker Used Anthropic's Claude AI to Steal South Korean Bank Data, CrowdStrike Reports¶
CrowdStrike identified a suspected 26-year-old Chinese attacker who used Anthropic's Claude AI and the ARTEX penetration-testing tool to compromise at least nine South Korean banks. The attacker queried Claude on selling stolen data and asked it to create a resume containing personal details.
- Suspect: 26-year-old, Guangdong province, Maoming location
- At least 9 South Korean banks targeted, late September to early October 2026
- Shinhan Bank: ~25,000 customer records compromised
- KB Kookmin Bank: 119 customer records leaked
Sources: Benzinga AI Web Searched
Competition¶
7 OpenAI gaining competitive ground on Anthropic in AI pricing war¶
OpenAI is gaining competitive ground on Anthropic in the AI pricing war. Chinese competitor DeepSeek's V4.1 Flash narrowed the performance gap to 3% on benchmarks, down from 9% in May, driving U.S. companies to cut prices as they pursue IPOs.
- DeepSeek V4.1 Flash LiveBench score: 81.1 vs Anthropic 83.4 (3% gap)
- Performance gap narrowed from 9% in May, 15% earlier in year
- Chinese models' lower costs pressuring U.S. pricing
- Both companies pursuing IPOs, requiring sustainable business models
Sources: Wall Street Journal AI Web Searched, The Straits Times AI Web Searched, WSJ RSS
Funding¶
6.5 Tech investors balk at Anthropic $3 trillion valuation¶
At a private poll, only three investors from a tech audience expressed willingness to purchase Anthropic shares at a $3 trillion valuation.
Sources: Business Insider RSS
AI Safety¶
5.5 Anthropic reduces safeguards for vetted security researchers testing Claude¶
Anthropic has reduced Claude's safety guardrails for authorized security researchers participating in vulnerability testing programs.
Sources: Dark Reading Web Search