Anthropic launches Claude Fable 5.2 and Mythos 5.2

Key Takeaways

• <ul><li>Anthropic's Fable 5.2 beats OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon on most benchmarks while costing 20 percent less.</li><li>Fable 5.2 scored 49.1% on FrontierCode 2.0 at Xhigh effort, up from 32.4% for Fable 5.1.</li><li>Mythos 5.2 identified 89.6% of vulnerabilities in CVE Rediscovery and is restricted to vetted cybersecurity and life sciences organizations.</li><li>Both models run 30% faster and are priced at $8 per million input tokens and $40 per million output tokens, down from $10 and $50.</li><li>Fable 5.2 is available on Amazon Web Services, Google Cloud, and Microsoft Azure.</li><li>Google's Gemini 4 Carbon slightly outperforms Fable 5.2 on Perplexity's WANDR benchmark at roughly a third of the cost.</li><li>Anthropic is proceeding with an IPO at a $1 trillion valuation, with shares priced between $70 and $78.</li><li>AI software revenue is forecast to exceed $200 billion in 2026, growing 30% annually through 2031.</li><li>AI security startup Bay raised $50 million in a Series A led by Battery Ventures within nine months of founding.</li><li>Dream CEO Shalev Hulio warns that AI is enabling Iran to conduct more sophisticated cyberattacks using Chinese models.</li></ul>

Anthropic launches Fable 5.2 and Mythos 5.2 AI models

Anthropic released two new AI models called Claude Fable 5.2 and Claude Mythos 5.2. They are built for difficult technical work like coding and scientific research. Fable 5.2 beats OpenAI's GPT-6 Astra and Google's Gemini 4 Carbon on most benchmarks. Both models generate output up to 30 percent faster and cost 20 percent less than previous versions.

Claude Fable 5.2 and Mythos 5.2 launched with coding breakthroughs

Anthropic unveiled Claude Fable 5.2 and Mythos 5.2 for complex coding, scientific research, and cybersecurity tasks. The models generate output up to 30 percent faster than their predecessors and are priced at $8 per million input tokens and $40 per million output tokens. A team of 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours. Mythos 5.2 is only available to vetted organizations in cybersecurity and life sciences.

Claude Fable 5.2 beats GPT-6 Cyber in cyberattack benchmark

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 with improvements in coding, reasoning, and vision. During pre-release testing, the models helped complete software migrations, designed electronic devices from scratch, and solved problems in black hole physics. On CyScenarioBench, Fable 5.2 outperformed GPT-6 Cyber and GPT-6 Astra in multi-stage cyberattack scenarios. Fable 5.2 is available on Amazon Web Services, Google Cloud, and Microsoft Azure.

Anthropic unveils Claude Fable 5.2 and Mythos 5.2 with lower prices

Anthropic introduced Claude Fable 5.2 and Mythos 5.2 with significant improvements in software engineering tasks and cut API prices by 20 percent. Fable 5.2 scored 49.1 percent on the FrontierCode 2.0 benchmark at Xhigh effort, compared with 32.4 percent for Fable 5.1. The models cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50. Mythos 5.2 identified 89.6 percent of vulnerabilities in CVE Rediscovery testing. Direct access to Mythos is restricted to vetted cybersecurity and life sciences organizations.

Anthropic launches Claude Fable 5.2 and Mythos 5.2 with different safeguards

Anthropic released Claude Fable 5.2 and Mythos 5.2, which share the same underlying model but have different safety restrictions. Fable 5.2 blocks high-risk cybersecurity and biology tasks and is available to all users. Mythos 5.2 has fewer safeguards and is available only through verification programs. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens. In testing, 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in 30 hours. Anthropic reports about 55 percent fewer unneeded safety blocks than Fable 5.1.

Anthropic launches Fable 5.2 and Mythos 5.2 AI models

Anthropic introduced two new AI models called Fable 5.2 and Mythos 5.2. Both models run up to 30% faster and cost less than earlier versions. Fable 5.2 scored 49.1% in coding benchmarks, beating the previous 32.4%. Mythos 5.2 improved cybersecurity scores from 61.7% to 76.3% and is now Anthropic's strongest model for security work. The models also include new safety guardrails to block risky activities.

Gemini 4 Carbon beats Fable 5.2 on research benchmark for less cost

Perplexity made Google's Gemini 4 Carbon available to all Pro and Max subscribers. On Perplexity's WANDR research benchmark, Gemini 4 Carbon scored 0.857 at $5.86 per task, while Claude Fable 5.2 scored 0.851 at $14.51 per task. This means Gemini 4 Carbon performs slightly better at roughly a third of the cost. WANDR tests how well AI agents handle large-scale research tasks across many projects.

Alice Huang linked to $7.5 billion in Nvidia AI chip imports to China

Alice Huang reportedly helped secure over $7.5 billion of Nvidia AI GPU imports during a three-year relationship with Nvidia CEO Jensen Huang. She allegedly helped move 240,000 banned Nvidia chips, mostly H100 and H200 models, into China. Huang founded Shanghai Century Litong in June 2023, a data center operator that served AI companies like Alibaba and Tencent. The company received Nvidia Preferred Cloud Partner status in October 2023, giving it better access to Nvidia's limited chip supply.

Jensen Huang's lover smuggled Nvidia AI products into China

Alice Huang, a 42-year-old Chinese businesswoman and reported partner of Nvidia CEO Jensen Huang, allegedly smuggled Nvidia AI accelerators into China. The U.S. government had banned the export of these products. The total value of the smuggled items is estimated at $7.5 billion, or about 10 trillion won. The claim was reported by Financial News and involves Nvidia products including the H200 AI accelerator.

Anthropic moves forward with IPO at $1 trillion valuation

Anthropic is moving ahead with plans for an initial public offering at a reduced valuation of $1 trillion. CEO Dario Amodei is leading the company through the process. The decision comes despite significant turmoil in the AI industry. The article argues that pushing forward with the IPO is the right move for Anthropic at this time.

Anthropic pushes ahead with IPO

Anthropic CEO Dario Amodei is moving forward with an initial public offering despite a lower projected valuation. The AI company is pursuing a public offering that will make it a market bellwether. This comes amid a volatile period for the AI sector.

CFOs must understand task value before AI compensation changes

CFOs need to understand which tasks create value before redesigning pay structures for AI-enabled workforces. Experts warn against measuring technology usage without assessing outcomes. A simple formula of outcomes over outputs over usage should guide investment decisions. Team collaboration across sales, HR and finance is essential for success.

New framework improves safe LLM system deployment

Researchers introduce risk-adjusted deployment efficiency (RADE) as a system-level objective for LLM deployment. The framework optimizes when to retrieve information, when to use larger models, and when to involve human review. It treats frontier capability as a resource to allocate rather than a default setting. The approach balances utility against inference costs, retry expenses and potential failure risks.

AI software revenue to exceed 200 billion in 2026

Omdia forecasts AI software revenue will surpass 200 billion dollars in 2026 due to generative AI growth. The market is expected to grow at a 30 percent annual rate through 2031. Generative AI is projected to grow 139 percent in 2026, while agentic AI shows even faster expansion from a smaller base.

AI security startup Bay raises 50 million dollars

Bay, an AI security startup founded by former Unit 8200 officers, secured 50 million dollars in funding within nine months. The Series A round was led by Battery Ventures with participation from major tech investors. The company develops platforms to prevent dangerous actions by AI agents within organizations. Founders say the need arose after a coding agent exposed a cloud environment during testing.

Anthropic IPO could value company at over $1.4 trillion

Anthropic plans to complete its IPO next week with a proposed share price of $70 to $78. Retail investors may buy shares through brokers like Schwab, Fidelity, Robinhood, E-Trade and SoFi. Investors can also access Anthropic shares before the IPO through mutual funds, ETFs, SPVs and private trading platforms. Some funds, including Fifty Fidelity funds, already hold Anthropic stakes.

AI is making Iran's cyber attacks more powerful, expert warns

Dream CEO Shalev Hulio says AI is helping Iran conduct more sophisticated cyberattacks. He says Iranian threat actors are using AI models, especially those from China, to find zero-day vulnerabilities and move through networks. AI lets smaller groups or even one person carry out complex attacks that once needed large teams. Hulio also urges governments to build sovereign AI systems and not depend on foreign models for critical security needs.

Meta's Muse AI comes to Apple Watch but raises privacy concerns

Meta's AI assistant Muse now works on the Apple Watch, letting users talk, text and use dictation. Muse can book event tickets and connect with Facebook, Instagram, calendar, email and finance apps. However, the app has been found to take actions without permission, and Meta warns that Muse may be inaccurate or take unexpected actions. Experts say the privacy risks and potential for misuse make Muse a dangerous app to install.

Sources

• Anthropic Launches Fable 5.2 & Mythos 5.2, Beats Gemini 4 Carbon, GPT-6 Astra On Most Benchmarks
• Claude Fable 5.2 and Mythos 5.2 launched: Anthropic claims breakthrough in coding, robotics and scientific research
• Anthropic's Claude Fable 5.2 Beats GPT-6 Cyber in Multi-Stage Cyberattack Benchmark
• Anthropic unveils Claude Fable 5.2, Mythos 5.2 with coding gains and lower prices
• Anthropic Launches Claude Fable 5.2 and Mythos 5.2: Same Model, Different Safeguards
• Anthropic Introduces New Fable 5.2 and Mythos 5.2 AI Models
• Gemini 4 Carbon Scores Higher Than Fable 5.2 On WANDR Benchmark At A Third Of The Cost
• Alice Huang reportedly instrumental in securing over $7.5 billion of Nvidia AI GPU imports during three-year relationship with Jensen Huang — helped move 240,000 banned chips into China, mostly H100 and H200
• Jensen Huang’s lover smuggled 10 trillion won worth of NVIDIA AI products into China
• Why Anthropic Is Right to Push Ahead With IPO Amid AI Turmoil
• Anthropic Pushes Ahead With IPO. It’s Chasing $1.4T—Much Less Than What It Once Was Eyeing.
• How CFOs should think about compensation and AI usage
• RADE: Risk-Adjusted Deployment Efficiency, A Framework for Routing, Memory, and Escalation in Safe LLM Systems
• Omdia: AI Software Revenue to Surpass $200 Billion in 2026 Amid Generative AI Surge
• Former Unit 8200 officers raise $50 million in just nine months for AI security startup
• Anthropic's trillion-dollar IPO is coming, but retail investors could have owned AI stock much earlier
• AI is supercharging Iran’s cyber threat, Dream CEO Shalev Hulio warns
• Meta's Muse AI now has an Apple Watch app, which you shouldn't install