2026.09.23 AI News Briefing — Claude Opus 5.5 and GPT-6 Price Cuts, and AI Safety Coordination

Opus 5.5 and GPT-6 cut prices in tandem, reigniting frontier-model competition. On the same day, three companies unveiled cybersecurity-focused AI safeguards, and the UN Security Council formally put AI risk on its agenda.

---

1. Anthropic Launches Claude Opus 5.5: 20% Price Cut, Usage Caps Scrapped

• Source: Benzinga | 2026.09.22
• What happened: Anthropic released Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, a 20% cut. Output speed is more than 30% faster, and the 5-hour usage caps have been eliminated.
• Why it matters: The price-cutting race is spreading to top-tier models.

2. OpenAI Launches GPT-6 Sol and Luna: Half the Price, Lower Error Rate

• Source: TechCrunch | 2026.09.22
• What happened: OpenAI unveiled the mainstream Sol and the ultra-light Luna as successors to the flagship Astra. Sol costs half as much as the previous generation ($2 input, $10 output) and its error rate is also halved.
• Why it matters: Unit prices keep falling on the mainstream models that carry most real-world traffic.

3. Google, Anthropic, and OpenAI Simultaneously Unveil Cybersecurity AI and Safeguards

• Source: The Hacker News | 2026.09.22
• What happened: Google announced the security model 'Gemini 3.8 Flash Cyber,' Anthropic the restricted-access model 'Claude Mythos 5.1' and enterprise safeguards (EFS), and OpenAI 'Astra,' which refuses 91.5% of jailbreak attempts.
• Why it matters: The three frontier labs strengthened their anti-misuse systems side by side on the same day.

4. UN Security Council Holds First Formal AI Risk Discussion, with Altman, Anthropic, and DeepSeek Attending

• Source: Business Standard | 2026.09.22
• What happened: At a Security Council meeting chaired by France on September 23, OpenAI and Anthropic along with China's DeepSeek and Moonshot will discuss the risk of losing control of AI. The U.S. and China are separately negotiating a mutual notification system for AI incidents.
• Why it matters: It is the first time the Security Council has brought U.S. and Chinese frontier companies to the same table.

5. Rebellions Raises KRW 640 Billion Pre-IPO as Capital Pours into 'K-NPU'

• Source: Nate News | 2026.09.21
• What happened: Korean AI chip (NPU) company Rebellions raised KRW 640 billion in pre-IPO funding. FuriosaAI also secured KRW 400 billion, as capital flows into domestic NPU and memory companies.
• Why it matters: Large sums are flowing into Korea's chip ecosystem as it tries to reduce dependence on Nvidia.

6. Baemin Previews AI and Delivery-Robot Tech as the Delivery-App AI Race Accelerates

• Source: Financial News | 2026.09.22
• What happened: Woowa Brothers will unveil AI and delivery-robot technology at 'WOOWACON 2026' in October. Kakao and Coupang Eats are also expanding their partnership on an in-KakaoTalk ordering and payment agent.
• Why it matters: Korean delivery platforms are expanding AI agents into live services.

---

Today's Insight

• The frontier-model price war has reached the very top tier, making 'unit price' as central a variable as performance.
• The three labs' safeguard announcements and the UN Security Council agenda item signal that building anti-misuse systems has become a shared industry task.