tags. We'll keep the structure. Let's go paragraph by paragraph. Original:
Anthropic shipped Claude Fable 5.1 on Tuesday, its newest flagship for coding and knowledge work, alongside a restricted sibling called Mythos 5.1. The release also tightens the screws on model copying: new developer accounts can no longer edit Claude's prior context in a multi-turn conversation while keeping the transcript of its thinking intact.
Translation:Anthropic于周二发布了Claude Fable 5.1，这是其面向编码和知识工作的最新旗舰模型，同时还有一个受限的姊妹模型Mythos 5.1。此次发布还加强了对模型复制的限制：新的开发者账户不能再在多轮对话中编辑Claude的先前上下文，同时保留其思考记录。
We need to ensure "tightens the screws" means "加强限制" or "收紧". "model copying" is "模型复制" or "模型抄袭". "distillation" is "蒸馏" but here it's about copying. We'll use "模型复制" for now. "transcript of its thinking" - "思考记录" or "思维链" but it's "transcript" so "记录". We'll say "思考记录". Next:The benchmark jump
->基准测试的飞跃
or "基准分数跃升". We'll do "基准测试的飞跃". Then:Anthropic says Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, more than double Fable 5's 24.7%. On Terminal-Bench 4.0 it hit 55.8%, ahead of Opus 5's 52.3% and OpenAI's GPT-5.6 Sol at 37.3%. Business workflow scores on AutomationBench nearly doubled to 31.4%.
Translation:Anthropic表示，Fable 5.1在Terminal-Bench-Science 0.1上得分52.6%，是Fable 5的24.7%的两倍多。在Terminal-Bench 4.0上，它达到55.8%，领先于Opus 5的52.3%和OpenAI的GPT-5.6 Sol的37.3%。在AutomationBench上的业务工作流得分几乎翻倍，达到31.4%。
We keep model names as is. Next:The knowledge cutoff moved to June 2026, five months later than Fable 5. That means the model has fresher information to draw on, which matters for coding tasks that depend on recent library changes and API updates.
Translation:知识截止日期移至2026年6月，比Fable 5晚了五个月。这意味着模型可以获取更新的信息，这对于依赖最新库更改和API更新的编码任务很重要。
"knowledge cutoff" is "知识截止日期" or "知识截止时间". We'll use "知识截止日期". Next:What it costs
->价格
or "成本". We'll do "价格".Pricing stays at $10 per million input tokens and $50 per million output tokens. Cache reads drop 75% to $0.25 per million. Anthropic claims typical workloads run 25% cheaper and complex agent tasks up to 45% cheaper with Fable 5.1.
Translation:定价保持不变，每百万输入令牌10美元，每百万输出令牌50美元。缓存读取价格下降75%，降至每百万0.25美元。Anthropic声称，使用Fable 5.1，典型工作负载的成本降低25%，复杂代理任务的成本降低高达45%。
"tokens" - we can use "令牌" or "标记". In Chinese AI context, "token" is often "令牌" or "标记". We'll use "令牌" for consistency. Next:Both models went live across Anthropic's platform, Amazon Bedrock, Google Cloud, and Microsoft Foundry. So the cheaper cache pricing applies everywhere those services are offered.
Translation:这两款模型已在Anthropic平台、Amazon Bedrock、Google Cloud和Microsoft Foundry上线。因此，更便宜的缓存定价适用于所有提供这些服务的地区。
Next:The distillation fight
->蒸馏之争
or "蒸馏打击". We'll do "蒸馏之争".Anthropic says it traced over 16 million Claude exchanges to distillation campaigns run through roughly 24,000 fake accounts. It named three Chinese labs — DeepSeek, Moonshot AI, and MiniMax — as responsible. Moonshot alone accounted for 3.4 million exchanges.
Translation:Anthropic表示，它追踪到超过1600万次Claude交互，这些交互来自通过约2.4万个虚假账户进行的蒸馏活动。它点名了三家中国实验室——DeepSeek、Moonshot AI和MiniMax——作为责任方。仅Moonshot就占了340万次交互。
"exchanges" - we can say "交互" or "对话". We'll use "交互". Next:White House science adviser Michael Kratsios accused Moonshot of copying Anthropic's flagship model to build its Kimi K3 system. Moonshot has not publicly responded.
Translation:白宫科学顾问Michael Kratsios指责Moonshot复制Anthropic的旗舰模型来构建其Kimi K3系统。Moonshot尚未公开回应。
Next:Distillation is the practice of using one model's outputs to train another. It's not new, but the scale here is striking. Sixteen million exchanges is a lot of raw material for a rival to work with.
Translation:蒸馏是一种使用一个模型的输出来训练另一个模型的做法。这并不新鲜，但这里的规模令人震惊。1600万次交互对于竞争对手来说是大量的原材料。
Next:
