[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-claude-opus-4-7-tokenizer-35pct-bill":3,"topics-all":37,"news-related-8bc2aee8-1a09-4e55-9965-d398cbeebab6":56},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":35,"view_count":36},"8bc2aee8-1a09-4e55-9965-d398cbeebab6","Claude Opus 4.7 新 tokenizer 背后的成本真相：标称价格不变，实际账单已悄然膨胀","Anthropic 于 4 月中旬发布 Claude Opus 4.7，带来了 SWE-bench Pro 64.3% 的编程能力新高和一个鲜少被注意的底层变化：新 tokenizer 将相同输入文本映射到更多 token，增幅在 1.0x 到 1.35x 之间，英文内容约高出 35%。这意味着按标称价格收费的 API，实际账单正在悄然膨胀。\n\nTokenization 是大模型处理文本的第一步，将输入文本切分成 token 序列。不同的 tokenizer 切分粒度不同，同一段话可能切出 1000 个 token，也可能切出 1350 个。Anthropic 此次更换了 Opus 系列的 tokenizer，但没有降低 per-token 定价——结果是用户每处理一批文本，实际消耗的 token 数变多了，而单价没变。\n\n按照 Opus 4.6 的定价，假设一段 5000 词的英文代码审查任务消耗 10,000 token，费用是 0.05 美元。换成 Opus 4.7，同一段任务可能消耗 13,500 token，费用升至 0.0675 美元，涨幅 35%。这个数字在单次调用中不起眼，但在日均百万 token 调用量级的生产系统中，月度账单差距可以轻松达到数千美元。\n\n更值得关注的是，这种增幅并不均匀。结构化代码、重复性日志、模板化文档的 token 增量普遍偏高；创意写作、对话类文本增量偏低。如果应用场景以代码为主，实际成本膨胀会显著高于官方宣称的平均","https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-opus-4-7","1fa87d30-d9f3-4752-b3be-0373933b3aaf",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",{"id":18,"name":19,"slug":19,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"fdc7f005-05cd-4810-8ff6-0197a8611f13","en","Opus 4.7's new tokenizer quietly inflates real-world bills","Anthropic released Claude Opus 4.7 in mid-April, bringing a new SWE-bench Pro record of 64.3% on coding capability and a rarely noticed underlying change: the new tokenizer maps the same input text to more tokens, with the increase ranging from 1.0× to 1.35× — about 35% higher for English content. This means that for APIs billed at the listed price, real bills are quietly inflating.\n\nTokenization is the first step in a large-model's text processing, slicing input text into a token sequence. Different tokenizers have different slicing granularity; the same passage might slice into 1,000 tokens or 1,350. Anthropic has changed the tokenizer for the Opus family, but hasn't lowered the per-token price — the result: every batch of text users process consumes more tokens, at the same unit price.\n\nAt Opus 4.6's pricing, suppose a 5,000-word English code-review task consumes 10,000 tokens — that costs $0.05. Switch to Opus 4.7 and the same task may consume 13,500 tokens, costing $0.0675 — a 35% jump. This number is invisible on a single call, but in production systems running millions of tokens per day, monthly bill differences easily reach thousands of dollars.\n\nMore notably, this increase is not uniform. Structured code, repetitive logs, and templated documents tend to see larger token increments; creative writing and dialogue text see smaller ones. If your application is code-heavy, real cost inflation will significantly exceed the official average of 35%.\n\nTokenizer changes have always been the elephant in the room. Claude Opus 4.7 is no isolated case — every historical tokenizer change has brought implicit cost reassessment, but most have stayed within 5%. This time, a 35% jump is large enough to invalidate carefully built cost models.\n\nFacing a tokenizer change, the most pragmatic approach is to re-measure token consumption with real input samples and build a cost baseline based on actual content. For high-frequency call scenarios, sample a typical batch, measure actual token differences using both Opus 4.6 and Opus 4.7 API endpoints, and derive the real inflation coefficient. Tokenization is foundational infrastructure for large models; its changes should not come as a surprise to users. The only reliable approach is to let your own data speak.","claude-opus-4-7-tokenizer-35pct-bill","2026-05-22T14:06:00Z","2026-05-22T22:06:02.761524Z","2026-08-19T02:08:40.142862Z",true,"agent","35%。\n\nTokenization 变更历来是房间里的大象。Claude Opus 4.7 并非孤例，历史上每次 tokenizer 更换都会带来隐性成本重估，只是幅度大多在 5% 以内。这一次 35% 的增幅已经高到足以让精细化成本模型失效。\n\n面对 tokenizer 变更，最务实的做法是用真实输入样本重新测量 token 消耗，建立基于实际内容的成本基准。对于高频调用场景，可以采样一个典型 batch，分别用 Opus 4.6 和 Opus 4.7 的 API 端点测量实际 token 差异，得出真实膨胀系数。Tokenizer 是大模型的基础设施，它的变化不应该成为用户的惊喜。唯一可靠的做法是用自己的数据说话。",134,[38,47],{"slug":39,"tag_slug":39,"title_zh":40,"title_en":41,"intro_zh":42,"intro_en":43,"id":44,"is_active":33,"created_at":45,"modified_at":46},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":48,"tag_slug":48,"title_zh":49,"title_en":50,"intro_zh":51,"intro_en":52,"id":53,"is_active":33,"created_at":54,"modified_at":55},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":57},[58,63,68,73,78,83],{"id":59,"title":60,"news_slug":61,"published_at":62},"6811f1f4-612b-4a99-824c-8678d2113177","Claude Opus 5 的真正卖点不是更强,而是 medium effort 这一档","claude-opus-5-medium-effort","2026-07-26T02:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"390c2437-4e4f-45ec-8270-67c5bfa4fa47","ChatGPT、Claude、Grok、Gemini 罕见同时下线,周四早晨全球 AI 集体失声","chatgpt-claude-grok-gemini-thursday-outage","2026-09-05T06:00:00+00:00",{"id":69,"title":70,"news_slug":71,"published_at":72},"f3d17d45-e1a8-4a1b-9449-6813aff06e49","Anthropic 让 Claude 自己修对齐:10 类失败全部见效,还超过人类研究员","claude-automated-alignment-researchers","2026-08-29T13:05:00+00:00",{"id":74,"title":75,"news_slug":76,"published_at":77},"39724847-fdc9-4199-ac46-311e7b49d385","Ramp 数据复盘 Fable 5:旗舰上市两月仅占企业 Anthropic 支出 11%,70 倍价差压住前沿模型溢价","ramp-data-fable-5-adoption-plateaus","2026-08-26T08:00:00+00:00",{"id":79,"title":80,"news_slug":81,"published_at":82},"e1724d68-bf0d-4b3f-8047-147796d5d52e","Ramp 8 月指数:Fable 5 企业份额停滞 11%,OpenAI 旗舰跑赢两倍","anthropic-fable-5-plateau-11-percent","2026-08-25T06:00:00+00:00",{"id":84,"title":85,"news_slug":86,"published_at":82},"1051d676-8ed9-4448-b0d5-8db4b844f41f","Claude Fable 5 上线两个月,为什么企业只把 11% 的账单花给最强模型","claude-fable-5-11-percent-anthropic-spend"]