{
  "id": 5073193,
  "title": "Frontier AI at a cost: what Anthropic’s Fable 5.1 means for the US-China model race",
  "url": "https://urgent.news/2026/09/02/frontier-ai-at-a-cost-what-anthropics-fable-5-1-means-for-the-us",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T11:30:06.000Z",
  "source": {
    "name": "SCMP Tech",
    "slug": "scmp-tech",
    "url": "https://www.scmp.com/tech/big-tech/article/3366125/frontier-ai-cost-what-anthropics-fable-51-means-us-china-model-race"
  },
  "original_language": "en",
  "account": "Anthropic's cutting-edge Claude Fable 5.1 model has surged ahead of Chinese counterparts in performance benchmarks, but a significant price difference highlights a widening gap between US and Chinese AI approaches. The American company unveiled Fable 5.1 alongside a restricted-access version, Mythos 5.1, stating it is now the world's most advanced model for software coding and complex knowledge work. Fable 5.1 earned the top spot in Artificial Analysis' Intelligence Index, scoring 66, a six-point lead over Chinese rivals Moonshot AI's Kimi K3 and Z.ai's GLM-5.3. Remarkably, Fable 5.1 also claimed first place on Vals AI's index for managing complex real-world tasks in finance, coding, and law. This benchmark data has fueled speculation about the United States' growing lead over China in the AI race. Experts have called Fable 5.1's capabilities \"insane,\" noting it surpassed claims that Chinese open-weight developers had closed the performance gap. However, the price gap signals a stark contrast between the two countries' strategies. While Anthropic reduced cache read prices by 75%, cutting user expenses by up to 45%, Fable 5.1 remains expensive at $3.69 per task, compared to $0.84 for Kimi K3 and $0.68 for GLM-5.3. This price difference suggests the market may naturally favor cheaper open models for routine tasks, while turning to Fable for more complex jobs. Chinese AI developers are also pushing ahead with performance advancements. Alibaba Group released Qwen3.8-Max-0902, which topped Arena AI's benchmark for front-end web development, outperforming Opus 5 and Kimi K3. Z.ai's upcoming GLM-6.0 aims to achieve recursive self-improvement, allowing the model to autonomously train its successor. Vals AI CEO Rayan Krishnan claimed Fable 5.1 exhibited signs of this capability in early testing. As the US-China AI competition intensifies, safety concerns are mounting. OpenAI teased its upcoming Astra model, which boasts cybersecurity capabilities to detect unknown vulnerabilities in well-protected systems without human intervention. However, safety researchers have raised alarms about Astra's \"recurrent depth\" architecture, which boosts performance but obscures the decision-making process, potentially impairing oversight. Redwood Research's Ryan Greenblatt warned the technique may be the \"single worst development for AI security and safety to date.\"",
  "summary": "Anthropic’s powerful new Claude Fable 5.1 model has widened its lead in performance benchmarks over Chinese rivals, even as budget-friendly open-weight models from China continue to gain commercial traction globally. The American company said on Wednesday that Fable 5.1, along with a restricted-access version called Mythos 5.1, was the world’s most advanced model for software coding and complex…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "South China Morning Post",
        "title": "Frontier AI at a cost: what Anthropic’s Fable 5.1 means for the US-China model race",
        "url": "https://urgent.news/2026/09/02/frontier-ai-at-a-cost-what-anthropics-fable-5-1-means-for-the-us-5074960",
        "published": "2026-09-02T11:30:06.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}