{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-23T06:40:50.084Z","headline":"前沿模型实际成本：tokenizer 差异导致隐性涨价","description":"同一份 TypeScript 文件在 GPT-5.x 上为 681 个 token，在 Claude 最新 tokenizer 下为 1，178 个，相差 1.73 倍。Anthropic 新 tokenizer 比旧版多产生约 30% 的 token，标价不变，构成隐性涨价。Claude Opus 4.6 与 Opus 4.8 标价同为 $5.00 / $25.00，但新 tokenizer 使同一代码的 token 数增加约 32%。Claude Sonnet 5 的 $2.00 / $10.00 为促销价，2026 年 8 月 31 日后恢复 $3.00 / $15.00，届时相同代码成本将比 Sonnet 4.6 高出约 32%。跨厂商对比中，Claude 新 tokenizer 在代码上比 GPT 多产生 1.50x 至 1.73x 的 token，TypeScript 差距最大。","url":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/","mainEntityOfPage":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/","datePublished":"2026-07-13T21:50:30.963Z","dateModified":"2026-07-13T21:50:30.963Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://playcode.io/blog/real-price-of-frontier-models","https://aihot.virxact.com/items/cmrjrg5aj0202bis4f9pzo8d3"],"canonicalUrl":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/","directAnswer":{"@type":"Answer","text":"文章称，同一内容在不同前沿模型的 tokenizer 下会产生不同 token 数，因此仅比较每百万 token 标价可能低估实际成本。其测试中，Claude 新 tokenizer 对英语和代码的计数增幅较明显。","url":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/","dateCreated":"2026-07-13T21:50:30.963Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"playcode.io source article","url":"https://playcode.io/blog/real-price-of-frontier-models","datePublished":"2026-07-13T21:50:30.963Z","provider":{"@type":"Organization","name":"playcode.io","url":"https://playcode.io/blog/real-price-of-frontier-models"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmrjrg5aj0202bis4f9pzo8d3","datePublished":"2026-07-13T21:50:30.963Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmrjrg5aj0202bis4f9pzo8d3"}}],"aggregationSource":"Hacker News 热门（buzzing.cc 中文翻译）","originalPublisher":{"name":"playcode.io","url":"https://playcode.io/blog/real-price-of-frontier-models"},"article":{"id":"cmrjrg5aj0202bis4f9pzo8d3","slug":"cmrjrg5aj0202bis4f9pzo8d3","url":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/","title":"前沿模型实际成本：tokenizer 差异导致隐性涨价","title_en":"前沿模型的实际价格。代币 * 价格，对吧？","summary":"同一份 TypeScript 文件在 GPT-5.x 上为 681 个 token，在 Claude 最新 tokenizer 下为 1，178 个，相差 1.73 倍。Anthropic 新 tokenizer 比旧版多产生约 30% 的 token，标价不变，构成隐性涨价。Claude Opus 4.6 与 Opus 4.8 标价同为 $5.00 / $25.00，但新 tokenizer 使同一代码的 token 数增加约 32%。Claude Sonnet 5 的 $2.00 / $10.00 为促销价，2026 年 8 月 31 日后恢复 $3.00 / $15.00，届时相同代码成本将比 Sonnet 4.6 高出约 32%。跨厂商对比中，Claude 新 tokenizer 在代码上比 GPT 多产生 1.50x 至 1.73x 的 token，TypeScript 差距最大。","source":"Hacker News 热门（buzzing.cc 中文翻译）","sourceUrl":"https://playcode.io/blog/real-price-of-frontier-models","aiHotUrl":"https://aihot.virxact.com/items/cmrjrg5aj0202bis4f9pzo8d3","publishedAt":"2026-07-13T21:50:30.963Z","category":"技巧观点","score":77,"selected":true,"articleBody":["We counted the same bytes under every frontier tokenizer, using each vendor's own counting endpoints, and cross-checked the counts against real paid requests. Below are the numbers and what they do to the prices on the rate cards.","A model's bill is two numbers multiplied together:","The pricing page shows the second number and treats the first as a constant. It is not a constant. It depends on the model's tokenizer, and tokenizers differ a lot between vendors. Two models can list the same \"$5.00 / 1M input tokens\" and produce meaningfully different bills for the same paragraph, because one of them turns that paragraph into more tokens. Since nobody publishes tokens-per-content numbers, we measured them.","We took 16 real fixtures: English prose, an HTML page, JavaScript, Python, TypeScript and Rust files, JSON tool schemas and tool results, Chinese chat and prose, symbol-heavy text, and our own agent system prompt. Each fixture was counted, byte for byte, with every model's production tokenizer:","GPT's o200k serves as the 1.00x reference throughout, mainly because it has been frozen and publicly documented for over two years, while Claude's tokenizer is the one that changed. DeepSeek and GLM are left out of the tables entirely: we only have rough characters-divided-by-four estimates for them, not real tokenizer counts, and this post is about measured numbers.","Claude Opus 4.6 and Opus 4.8 have the same $5.00 / $25.00 list price. What changed between them is the tokenizer. Sonnet 4.6 and Opus 4.6 use the old one; Sonnet 5, Opus 4.8, and Fable 5 use the new one. The table counts the same bytes with both, on Anthropic's own endpoint:","Weight those rows the way a real agent request is composed, which is mostly English system prompt, tool schemas, code, and JSON, and the new tokenizer comes out around +32% per request. The Chinese row barely moved, so the inflation is concentrated in English and code.","Sonnet 5 launched at $2.00 / $10.00, down from Sonnet 4.6's $3.00 / $15.00, which looked like a price cut. That is an intro price, and it ends August 31, 2026. While it lasts, the lower rate slightly more than covers the extra tokens, so Sonnet 5 works out a little cheaper than 4.6 for the same code. From September 1 the price returns to $3.00 / $15.00, the extra tokens remain, and the same work will cost about a third more than it did on Sonnet 4.6 at the same list price.","count_tokens is a prediction, so we also sent real paid requests with max_tokens: 1 and read usage.input_tokens , which is what invoices are based on. For the same content, Opus 4.6 billed 2,541 input tokens and Opus 4.8 billed 3,191 , each matching its predicted count exactly. We ran the same check on Fable 5, the most expensive model in the lineup, and it billed 3,191 as well, identical to Opus 4.8. So Fable uses the same new tokenizer and there is no extra per-token markup hidden behind its higher list price. The whole verification cost about $0.08.","The cross-vendor table uses GPT's o200k as the 1.00x reference. Every cell is that model's token count for the identical file divided by GPT's, so 1.20x means 20% more tokens than GPT. Claude's new and old tokenizers are shown side by side:","The code rows sit well above the prose rows: TypeScript at 1.73x , Rust at 1.58x, JavaScript at 1.52x, Python at 1.50x, against 1.40x for English prose. Code is most of what a coding agent processes, so for that workload the 1.50-1.73x band is the relevant one.","Why is TypeScript the worst case? Because o200k is unusually efficient on it: about 4.24 characters per token, which looks like the result of training on a lot of web JavaScript and TypeScript, where camelCase identifiers and JSX patterns compress into single tokens. On Rust its efficiency drops to about 3.51 characters per token. Claude's tokenizer is roughly equally dense on both languages, so the gap is widest exactly where GPT is strongest.","Chinese behaves differently. Claude sits around 1.45-1.55x above GPT with both the old and the new tokenizer (435 vs 433 tokens against GPT's 300 on the prose fixture), so this is a long-standing property of the Claude family on CJK text, not something the new tokenizer introduced. Gemini is actually more efficient than GPT here, at 256 tokens. Which tokenizer costs you more depends on what you write.","Multiply the list price by the measured divergence and you get an effective price for processing the same work. Divergence here is the blended multiplier for a typical English coding request, normalized to GPT's o200k :","A few rows are worth a second look. Opus 4.6 and 4.8 share a list price but differ by about 32% in effective price. GPT-5.5 and GPT-5.6 Sol share the tokenizer, so their identical list prices really are identical in effect. Gemini 3 Flash runs a slightly heavier tokenizer than GPT and still remains the cheapest option by a wide margin.","For an independent data point: Ploy published a production migration to GPT-5.6 Sol：https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6 this week and reported 1.70M input tokens against Claude Opus 4.8's 2.60M for the same builds, about 35% fewer. That is a whole-task bill rather than a tokenizer probe, so it also folds in model verbosity, but it points the same way.","Everything above measures one thing: how many input tokens identical bytes become. A full agent task adds more variables on top, and they are big ones. How many output and thinking tokens does the model spend to reach the same result? How much context does the harness load per step? How often does it call tools or spawn subagents? How does the provider price cache reads and writes?","Two consequences are worth spelling out. First, cache traffic is billed per token too, so a tokenizer that produces 32% more tokens also makes every cache write and every cache read about 32% more expensive, and on long agent sessions cache reads are most of the bill. Second, whole-task costs can diverge far more than 1.73x in either direction once verbosity and thinking are folded in. When people report that one model \"uses 2-4x the tokens\" of another on agent work, that can be true for their setup even though the pure input tokenization gap in our fixtures never exceeded 1.73x. The two numbers measure different layers.","By content type, the input-side range we measured for Claude's new tokenizer against GPT's o200k is: prose, HTML, and JSON at 1.36-1.42x; code at 1.50-1.73x with TypeScript on top; Chinese and symbol-heavy text at 1.44-1.53x. We put TypeScript in the title because it is both the top of the range and the thing a coding agent processes all day, not because the whole world is 1.73x.","We also measured the per-task layer directly, in a follow-up: the same nine models given one identical drawing task, one shot each, with every attempt priced from the providers' own usage numbers. The same drawing ranged from $0.004 to $0.80 depending on model and reasoning effort, and maximum effort made three models fail to finish at all. That experiment is here: The Pelican Benchmark Is Saturated. We Made 9 AI Models Draw a MacBook Pro Instead：/blog/macbook-svg-benchmark.","None of this makes one model universally right. GPT-5.x is the token-lean choice on English and code, Gemini 3 Flash is remarkably cheap in effect, and Claude models earn their place on quality even when they cost more tokens to run. Just make sure the price you compare is the one you actually pay, after the tokenizer.","Sources: Anthropic pricing (anthropic.com/pricing：https://www.anthropic.com/pricing), OpenAI pricing (platform.openai.com/docs/pricing：https://platform.openai.com/docs/pricing), Google Gemini API pricing (ai.google.dev：https://ai.google.dev/gemini-api/docs/pricing), xAI pricing (docs.x.ai：https://docs.x.ai). Token counts come from Anthropic's count_tokens endpoint, OpenAI's o200k_base (verified against live API usage), and the Google and xAI count endpoints. No text was generated to produce these counts.","The measurements are ours; the prose was drafted with AI assistance and edited by a human. Updated 2026-07-14 with a TL;DR, tighter wording, and a section on what input tokenization does not capture, based on reader feedback.","Playcode keeps every one of these models one click apart, so you can run the same prompt on two of them and compare the result that matters, the app it builds, instead of arguing about a sticker. Try it at playcode.io：https://playcode.io.","The follow-up: same models, one unforgiving drawing task, real per-task costs.","A long-form comparison of the three big AI families on real coding tasks.","What each of the three new GPT-5.6 tiers costs and when to reach for it.","The best UI and design quality we have tested, and when the premium is worth it.","We'd love to hear from you! Chat with us or send us an email.","© 2026 Playcode. Headquartered in Europe."],"articleImages":[{"sourceUrl":"https://playcode.io/marketing/assets/static/hero.CmWpi4qz.webp","alt":"Bar chart: the same 2,888-character TypeScript file becomes 681 tokens on GPT-5.x, 718 on Grok 4.5, 788 on Gemini 3 Flash, 898 on Claude's old tokenizer, and 1,178 on Claude's new tokenizer","afterParagraph":0,"url":"/media/articles/cmrjrg5aj0202bis4f9pzo8d3/fa5276067ea7dafc.webp"}],"mediaStatus":"ok","articleBodyZh":["我们在每个前沿分词器下统计了相同的字节，使用每个厂商自己的计数端点，并将计数与实际付费请求交叉核对。下面是这些数字以及它们对费率表价格的影响。","一个模型的账单是两个数字相乘的结果：","定价页面显示第二个数字，并将第一个数字视为常数。但它不是常数。它取决于模型的分词器，而分词器在不同厂商之间差异很大。两个模型可以列出相同的“$5.00 / 1M 输入令牌”，但针对同一段文字产生显著不同的账单，因为其中一个模型将这段文字分成了更多的令牌。由于没有人公开每内容的令牌数量，我们自己进行了测量。","我们选择了16个实际样例：英文散文、一个HTML页面、JavaScript、Python、TypeScript和Rust文件、JSON工具模式和工具结果、中文聊天和散文、符号密集的文本，以及我们自己的代理系统提示。每个样例都用每个模型的生产分词器按字节计数：","GPT的o200k作为1.00x参考，主要是因为它已经冻结并公开记录超过两年，而Claude的分词器进行了更改。DeepSeek和GLM完全不包含在表中：我们只有粗略的字符除以四的估算值，没有真实的分词器计数，而本文关注的是测量后的数字。","Claude Opus 4.6和Opus 4.8的列表价格都是$5.00 / $25.00。它们之间变化的是分词器。Sonnet 4.6和Opus 4.6使用旧分词器；Sonnet 5、Opus 4.8和Fable 5使用新分词器。下表在Anthropic自己的端点上用两种分词器统计相同的字节：","根据真实代理请求的组成方式对这些行加权，大多数是英文系统提示、工具模式、代码和JSON，新分词器每次请求大约增加32%。中文行几乎没有变化，所以价格上涨主要集中在英文和代码上。","Sonnet 5 的起始价格为 $2.00 / $10.00，低于 Sonnet 4.6 的 $3.00 / $15.00，这看起来像是降价。这是一个引导价格，截止日期为 2026 年 8 月 31 日。在此期间，较低的费率稍微多于覆盖额外的 tokens，因此对于相同的代码，Sonnet 5 比 4.6 稍微便宜一些。从 9 月 1 日起，价格将恢复到 $3.00 / $15.00，额外的 tokens 保持不变，相同的工作在 Sonnet 4.6 的相同标价下将比以前贵大约三分之一。","count_tokens 是一种预测，因此我们还发送了真实的付费请求，使用 max_tokens: 1 并读取 usage.input_tokens，这是发票依据。对于相同的内容，Opus 4.6 计费 2,541 个输入 tokens，Opus 4.8 计费 3,191 个，每个都完全匹配其预测数量。我们对 Fable 5 也进行了相同的检查，这是产品线中最昂贵的模型，它也计费 3,191 个，和 Opus 4.8 完全一致。因此，Fable 使用相同的新分词器，其更高的标价背后没有隐藏额外的每 token 加价。整个验证成本大约为 $0.08。","跨供应商表以 GPT 的 o200k 作为 1.00x 参考。每个单元格是该模型针对相同文件的 token 数除以 GPT 的 token 数，因此 1.20x 意味着比 GPT 多 20% 的 tokens。Claude 的新旧分词器并列显示：","代码行明显高于散文行：TypeScript 为 1.73x，Rust 为 1.58x，JavaScript 为 1.52x，Python 为 1.50x，而英文散文为 1.40x。代码占编码代理处理工作的大部分，因此在该工作负载下，1.50-1.73x 范围是相关的。","为什么 TypeScript 是最坏情况？因为 o200k 在其上效率异常高：每个 token 约 4.24 个字符，这看起来像是训练了大量网页 JavaScript 和 TypeScript 的结果，其中的 camelCase 标识符和 JSX 模式压缩成单个 token。在 Rust 上，它的效率下降到每个 token 约 3.51 个字符。Claude 的分词器在两种语言上的密度大致相等，因此差距恰好在 GPT 最强的地方最大。","中文的表现有所不同。Claude 的表现大约比 GPT 高出 1.45-1.55 倍，无论是使用旧的还是新的分词器（在散文测试中分别为 435 对 433 个标记，而 GPT 为 300 个标记），因此这是 Claude 系列在 CJK 文本上的一个长期特性，而不是新分词器引入的。Gemini 在这里实际上比 GPT 更高效，只有 256 个标记。哪个分词器成本更高，取决于你写的内容。","将标价乘以测得的差异，就可以获得处理相同工作的有效价格。这里的差异是典型英语编码请求的混合乘数，并已归一化为 GPT 的 o200k：","有几行数据值得再看一次。Opus 4.6 和 4.8 具有相同的标价，但有效价格相差约 32%。GPT-5.5 和 GPT-5.6 Sol 使用相同的分词器，因此它们标价相同，实际效果也真的相同。Gemini 3 Flash 的分词器略重于 GPT，但仍然在很大幅度上保持最便宜的选项。","作为一个独立的数据点：Ploy 本周发布了向 GPT-5.6 Sol 的生产迁移：https://ploy.ai/blog/migrating-a-production-ai-agent-to-gpt-5-6 并报告在相同构建下输入标记为 1.70M，而 Claude Opus 4.8 为 2.60M，减少约 35%。这是整个任务的计费，而不是分词器测试，因此还包括模型的冗长性，但方向一致。","上面的一切都测量了一件事：相同字节会变成多少输入标记。完整的代理任务增加了更多变量，而且这些都是重要的变量。模型为达到相同结果会花费多少输出和思考标记？每步上下文加载了多少？调用工具或生成子代理的频率如何？供应商如何定价缓存的读写？","有两个后果值得说明。首先，缓存流量也按 token 收费，因此产生 32% 更多 token 的分词器会使每次缓存写入和每次缓存读取的成本增加约 32%，在长时间的智能代理会话中，缓存读取占据账单的大部分。其次，一旦加入冗长性和思考环节，整个任务的成本可能在任意方向上偏离远超过 1.73 倍。当人们报告某个模型在代理工作中“使用的 token 是另一个模型的 2-4 倍”时，这可能在他们的设置中是真的，即使我们固定场景中的纯输入分词差距从未超过 1.73 倍。这两个数字测量的是不同层面。","按内容类型，我们测量了 Claude 新分词器相较于 GPT 的 o200k 的输入端范围：散文、HTML 和 JSON 为 1.36-1.42 倍；代码为 1.50-1.73 倍（TypeScript 最突出）；中文和符号密集文本为 1.44-1.53 倍。我们在标题中提到 TypeScript，是因为它既处于范围的上限，又是编码代理整天处理的内容，而不是因为整个世界都是 1.73 倍。","我们还直接测量了每个任务层，在后续实验中：同样九个模型执行了一个相同的绘图任务，每个模型尝试一次，每次尝试的价格基于提供商的实际使用数据。相同绘图任务的成本根据模型和推理努力程度从 0.004 美元到 0.80 美元不等，最大努力甚至导致三个模型完全未能完成。该实验在这里：The Pelican Benchmark Is Saturated. We Made 9 AI Models Draw a MacBook Pro Instead：/blog/macbook-svg-benchmark。","这些都不能说明某个模型在所有情况下一定正确。GPT-5.x 是处理英文和代码时 token 使用最少的选择，Gemini 3 Flash 在实际效果上显著便宜，而 Claude 模型即使消耗更多 token 运行，也凭借质量赢得市场地位。只要确保你比较的价格是你实际支付的价格，计算了分词器之后。","数据来源：Anthropic 定价（anthropic.com/pricing：https://www.anthropic.com/pricing）、OpenAI 定价（platform.openai.com/docs/pricing：https://platform.openai.com/docs/pricing）、Google Gemini API 定价（ai.google.dev：https://ai.google.dev/gemini-api/docs/pricing）、xAI 定价（docs.x.ai：https://docs.x.ai）。Token 数量来自 Anthropic 的 count_tokens 接口、OpenAI 的 o200k_base（已与实时 API 使用验证）、以及 Google 和 xAI 的计数接口。为计算这些数量未生成任何文本。","这些测量是我们的；文本由 AI 辅助起草，由人类编辑。于 2026-07-14 更新，增加了 TL;DR、更加简洁的表述，以及基于读者反馈增加了有关输入分词无法捕捉内容的部分。","Playcode 将这些模型保持在一个点击之内，因此你可以在两个模型上运行相同的提示，并比较重要的结果——它构建的应用，而不是关于贴纸争论。试试它：playcode.io：https://playcode.io。","后续：相同模型，一个苛刻的绘图任务，每个任务的真实成本。","对三大家庭 AI 在真实编码任务上的长篇比较。","三种新 GPT-5.6 层级的成本及使用时机。","我们测试过的最佳界面和设计质量，以及何时值得付费使用。","我们希望听到你的意见！与我们聊天或发送电子邮件。","© 2026 Playcode。总部设在欧洲。"],"translationStatus":"translated","bodyOrigin":"source-page","editorial":{"summary":"文章称，同一内容在不同前沿模型的 tokenizer 下会产生不同 token 数，因此仅比较每百万 token 标价可能低估实际成本。其测试中，Claude 新 tokenizer 对英语和代码的计数增幅较明显。","background":"作者使用各厂商计数接口，对16组英语、代码、JSON、中文及系统提示等相同字节内容进行统计，并以真实付费请求交叉核对。GPT 的 o200k 被设为1.00倍参考，缺少实测计数的模型未纳入表格。","viewpoint":"Aioga 判断，这篇文章的核心价值是提醒采购方将“单位 token 价格”和“同一任务产生的 token 数”同时纳入评估。不过，约32%的增幅来自作者对代理请求结构的加权结果，不宜直接外推到所有负载。","implications":"对以英语系统提示、工具定义、代码和 JSON 为主的代理任务，Claude 新 tokenizer 可能削弱标价下降带来的成本优势。中文样本变化较小，说明成本影响可能随语言、内容类型和请求组成而显著不同。","nextStep":"值得关注的是，团队可选取自身真实提示、代码、工具调用结果和输出样本，通过厂商正式计数接口测算 token，并结合输入与输出费率比较单次任务成本；同时应在促销价结束前后重新核算。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-07-22T19:12:16.491Z","sourceHash":"d279b24efb4c1f7c","review":{"approved":true,"groundedness":97,"clarity":94,"duplicationRisk":18,"blockingIssues":[],"notes":["候选内容准确区分了来源事实与“Aioga 判断”，并对约32%的加权结果增加了适用范围限制，避免将其误导性地外推至所有负载。","“Claude 新 tokenizer 可能削弱标价下降带来的成本优势”表述较为审慎；来源进一步指出，Sonnet 5 促销期间的降价略高于额外 token 成本，而促销结束并恢复原价后，相同工作成本将比 Sonnet 4.6 高约三分之一。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["技巧观点","Hacker News 热门（buzzing.cc 中文翻译）"],"translations":{"zh-CN":{"title":"前沿模型实际成本：tokenizer 差异导致隐性涨价","summary":"同一份 TypeScript 文件在 GPT-5.x 上为 681 个 token，在 Claude 最新 tokenizer 下为 1，178 个，相差 1.73 倍。Anthropic 新 tokenizer 比旧版多产生约 30% 的 token，标价不变，构成隐性涨价。Claude Opus 4.6 与 Opus 4.8 标价同为 $5.00 / $25.00，但新 tokenizer 使同一代码的 token 数增加约 32%。Claude Sonnet 5 的 $2.00 / $10.00 为促销价，2026 年 8 月 31 日后恢复 $3.00 / $15.00，届时相同代码成本将比 Sonnet 4.6 高出约 32%。跨厂商对比中，Claude 新 tokenizer 在代码上比 GPT 多产生 1.50x 至 1.73x 的 token，TypeScript 差距最大。","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"前沿模型实际成本：tokenizer 差异导致隐性涨价 - Aioga AI资讯","description":"同一份 TypeScript 文件在 GPT-5.x 上为 681 个 token，在 Claude 最新 tokenizer 下为 1，178 个，相差 1.73 倍。Anthropic 新 tokenizer 比旧版多产生约 30% 的 token，标价不变，构成隐性涨价。Claude Opus 4.6 与 Opus 4.8 标价同为 $5.00 / $","url":"https://www.aioga.com/news/cmrjrg5aj0202bis4f9pzo8d3/"},"en":{"title":"Frontier model actual cost: Tokenizer differences lead to hidden price increases","summary":"The same TypeScript file contains 681 tokens on GPT-5.x, and 1,178 tokens on Claude's latest tokenizer—a difference of 1.73 times. Anthropic's new tokenizer generates about 30% more tokens than the old version, with the price price unchanged, resulting in a hidden price increase. Claude Opus 4.6 and Opus 4.8 are both priced at $5.00 / $25.00, but the new tokenizer increases the number of tokens in the same code by about 32%. Claude Sonnet 5's $2.00 / $10.00 is a promotional price, and after August 31, 2026, it will revert to $3.00 / $15.00, at which point the same code will cost about 32% higher than Sonnet 4.6. In cross-vendor comparisons, Claude's new tokenizer generates 1.50x to 1.73x more tokens than GPT in code, with TypeScript showing the largest gap.","category":"Insights","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Frontier model actual cost: Tokenizer differences lead to hidden price increases - Aioga AI News","description":"The same TypeScript file contains 681 tokens on GPT-5.x, and 1,178 tokens on Claude's latest tokenizer—a difference of 1.73 times. Anthropic's new tokenizer generates about 30% mor","url":"https://www.aioga.com/en/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:50:58.971Z"},"ja":{"title":"フロンティアモデルの実際のコスト:トークナイザーの違いが隠れた価格上昇を引き起こします","summary":"同じTypeScriptファイルにはGPT-5.xで681トークン、Claudeの最新のトークナイザーでは1,178トークンが含まれており、1.73倍の差があります。 Anthropicの新しいトークナイザーは旧バージョンより約30%多くトークンを生成し、価格は変わらず隠れた価格上昇が見られます。 Claude Opus 4.6とOpus 4.8はどちらも$5.00 / $25.00で販売されていますが、新しいトークナイザーにより同じコード内のトークン数が約32%増加します。 『クロード・ソネット5』の$2.00 / $10.00はプロモーション価格であり、2026年8月31日以降は$3.00 / $15.00に戻り、その時点で同じコードはソネット4.6より約32%高くなります。 ベンダー間比較では、Claudeの新しいトークナイザーはGPTのコード上で1.50倍から1.73倍のトークンを生成しており、TypeScriptが最大の差を示しています。","category":"ヒントと視点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"フロンティアモデルの実際のコスト:トークナイザーの違いが隠れた価格上昇を引き起こします - Aioga AIニュース","description":"同じTypeScriptファイルにはGPT-5.xで681トークン、Claudeの最新のトークナイザーでは1,178トークンが含まれており、1.73倍の差があります。 Anthropicの新しいトークナイザーは旧バージョンより約30%多くトークンを生成し、価格は変わらず隠れた価格上昇が見られます。 Claude Opus 4.6とOpus 4.8はどちらも$","url":"https://www.aioga.com/ja/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:50:59.778Z"},"ko":{"title":"프런티어 모델 실제 비용: 토큰화 차이가 숨겨진 가격 상승을 초래합니다","summary":"동일한 TypeScript 파일에는 GPT-5.x에서 681개의 토큰, Claude의 최신 토큰라이저에서는 1,178개의 토큰이 포함되어 있어 1.73배 차이가 있습니다. Anthropic의 새로운 토큰라이저는 이전 버전보다 약 30% 더 많은 토큰을 생성하며, 가격 가격은 변하지 않아 숨겨진 가격 상승이 발생합니다. 클로드 Opus 4.6과 Opus 4.8 모두 $5.00 / $25.00으로 가격이 책정되어 있지만, 새로운 토큰 제작 덕분에 동일한 코드 내 토큰 수가 약 32% 증가합니다. 클로드 소네트 5의 $2.00 / $10.00은 프로모션 가격이며, 2026년 8월 31일 이후에는 $3.00 / $15.00으로 되돌아갑니다. 이 시점부터 같은 코드는 소네트 4.6보다 약 32% 더 비싸게 됩니다. 벤더 간 비교에서 Claude의 새로운 토큰라이저는 GPT보다 1.50배에서 1.73배 더 많은 토큰을 생성하며, TypeScript가 가장 큰 격차를 보입니다.","category":"인사이트","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"프런티어 모델 실제 비용: 토큰화 차이가 숨겨진 가격 상승을 초래합니다 - Aioga AI 뉴스","description":"동일한 TypeScript 파일에는 GPT-5.x에서 681개의 토큰, Claude의 최신 토큰라이저에서는 1,178개의 토큰이 포함되어 있어 1.73배 차이가 있습니다. Anthropic의 새로운 토큰라이저는 이전 버전보다 약 30% 더 많은 토큰을 생성하며, 가격 가격은 변하지 않아 숨겨진 가격 상승이 발생합니다. 클","url":"https://www.aioga.com/ko/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:50:59.297Z"},"es":{"title":"Coste real del modelo Frontier: Las diferencias en los tokenizadores conducen a aumentos ocultos de precio","summary":"El mismo archivo TypeScript contiene 681 tokens en GPT-5.x y 1.178 tokens en el último tokenizador de Claude—una diferencia de 1,73 veces. El nuevo tokenizador de Anthropic genera aproximadamente un 30% más de tokens que la versión anterior, manteniendo el precio sin cambios, lo que resulta en un aumento oculto de precio. Claude Opus 4.6 y Opus 4.8 tienen ambos un precio de 5,00 $ / 25,00 $, pero el nuevo tokenizador aumenta el número de tokens en el mismo código en aproximadamente un 32%. Los 2,00 $ / 10,00 $ de Claude Sonnet 5 son un precio promocional, y después del 31 de agosto de 2026, volverá a ser 3,00 $ / 15,00 $, momento en el que el mismo código costará aproximadamente un 32% más que el Sonet 4,6. En comparaciones entre proveedores, el nuevo tokenizador de Claude genera entre 1,50 y 1,73 veces más tokens que GPT en código, siendo TypeScript la mayor diferencia.","category":"Ideas","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Coste real del modelo Frontier: Las diferencias en los tokenizadores conducen a aumentos ocultos de precio - Aioga Noticias de IA","description":"El mismo archivo TypeScript contiene 681 tokens en GPT-5.x y 1.178 tokens en el último tokenizador de Claude—una diferencia de 1,73 veces. El nuevo tokenizador de Anthropic genera ","url":"https://www.aioga.com/es/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:50:59.484Z"},"fr":{"title":"Modèle frontière coût réel : les différences de tokenizer conduisent à des augmentations cachées des prix","summary":"Le même fichier TypeScript contient 681 jetons sur GPT-5.x, et 1 178 jetons sur le dernier tokeniseur de Claude — soit une différence de 1,73 fois. Le nouveau tokeniseur d’Anthropic génère environ 30 % de jetons en plus que l’ancienne version, le prix reste inchangé, ce qui entraîne une augmentation cachée du prix. Claude Opus 4.6 et Opus 4.8 sont tous deux proposés à 5,00 $ / 25,00 $, mais le nouveau tokeniseur augmente le nombre de jetons dans le même code d’environ 32 %. Le prix promotionnel de Claude Sonnet 5 est de 2,00 $ / 10,00 $, et après le 31 août 2026, il reviendra à 3,00 $ / 15,00 $, auquel cas le même code coûtera environ 32 % plus cher que Sonnet 4,6. Dans les comparaisons entre fournisseurs inter-fournisseurs, le nouveau tokeniseur de Claude génère de 1,50 à 1,73x plus de jetons que GPT en code, TypeScript affichant le plus grand écart.","category":"Analyses","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Modèle frontière coût réel : les différences de tokenizer conduisent à des augmentations cachées des prix - Aioga Actualités IA","description":"Le même fichier TypeScript contient 681 jetons sur GPT-5.x, et 1 178 jetons sur le dernier tokeniseur de Claude — soit une différence de 1,73 fois. Le nouveau tokeniseur d’Anthropi","url":"https://www.aioga.com/fr/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:50:59.449Z"},"de":{"title":"Frontier-Modell tatsächliche Kosten: Unterschiede in Tokenizern führen zu versteckten Preiserhöhungen","summary":"Die gleiche TypeScript-Datei enthält 681 Token auf GPT-5.x und 1.178 Token auf Claudes neuestem Tokenizer – ein Unterschied von 1,73 Mal. Anthropics neuer Tokenizer generiert etwa 30 % mehr Token als die alte Version, wobei der Preispreis unverändert bleibt, was zu einer versteckten Preiserhöhung führt. Claude Opus 4.6 und Opus 4.8 kosten beide 5,00 $ / 25,00 $, aber der neue Tokenizer erhöht die Anzahl der Token im selben Code um etwa 32 %. Claude Sonnet 5 ist ein Aktionspreis, der nach dem 31. August 2026 auf 3,00 $ / 15,00 $ zurückfällt, wobei derselbe Code etwa 32 % teurer ist als Sonnet 4.6. Im vendorübergreifenden Vergleich erzeugt Claudes neuer Tokenizer 1,50- bis 1,73-mal mehr Tokens als GPT im Code, wobei TypeScript die größte Lücke zeigt.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Frontier-Modell tatsächliche Kosten: Unterschiede in Tokenizern führen zu versteckten Preiserhöhungen - Aioga KI-News","description":"Die gleiche TypeScript-Datei enthält 681 Token auf GPT-5.x und 1.178 Token auf Claudes neuestem Tokenizer – ein Unterschied von 1,73 Mal. Anthropics neuer Tokenizer generiert etwa ","url":"https://www.aioga.com/de/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:02.772Z"},"pt-BR":{"title":"Custo real do modelo Frontier: Diferenças entre tokenizadores levam a aumentos ocultos de preço","summary":"O mesmo arquivo TypeScript contém 681 tokens no GPT-5.x e 1.178 tokens no tokenizador mais recente do Claude — uma diferença de 1,73 vezes. O novo tokenizador da Anthropic gera cerca de 30% mais tokens do que a versão antiga, mantendo o preço inalterado, resultando em um aumento oculto de preço. Claude Opus 4.6 e Opus 4.8 têm preços de $5,00 / $25,00, mas o novo tokenizador aumenta o número de tokens no mesmo código em cerca de 32%. O preço promocional de $2,00 / $10,00 do Claude Sonnet 5, e após 31 de agosto de 2026, voltará a ser $3,00 / $15,00, momento em que o mesmo código custará cerca de 32% mais que o Sonet 4,6. Em comparações entre fornecedores, o novo tokenizador do Claude gera de 1,50x a 1,73x mais tokens que o GPT em código, com o TypeScript mostrando a maior diferença.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Custo real do modelo Frontier: Diferenças entre tokenizadores levam a aumentos ocultos de preço - Aioga Notícias de IA","description":"O mesmo arquivo TypeScript contém 681 tokens no GPT-5.x e 1.178 tokens no tokenizador mais recente do Claude — uma diferença de 1,73 vezes. O novo tokenizador da Anthropic gera cer","url":"https://www.aioga.com/pt-BR/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:04.065Z"},"ru":{"title":"Реальная стоимость модели Frontier: различия токенайзеров приводят к скрытому росту цен","summary":"Тот же файл TypeScript содержит 681 токен на GPT-5.x и 1 178 токена на последнем токенайзере Claude — разница в 1,73 раза. Новый токенайзер Anthropic генерирует примерно на 30% больше токенов, чем старая версия, при этом цена не меняется, что приводит к скрытому росту цены. Claude Opus 4.6 и Opus 4.8 оцениваются по $5.00 / $25.00, но новый токенайзер увеличивает количество токенов в одном и том же коде примерно на 32%. Цена Claude Sonnet 5 в $2.00 / $10.00 — это промо-цена, а после 31 августа 2026 года цена вернётся к $3.00 / $15.00, после чего тот же код будет стоить примерно на 32% дороже, чем Sonnet 4.6. При сравнении между поставщиками новый токенайзер Claude генерирует в коде в 1,50–1,73 раза больше токенов, чем GPT, при этом TypeScript показывает наибольший разрыв.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Реальная стоимость модели Frontier: различия токенайзеров приводят к скрытому росту цен - Aioga Новости ИИ","description":"Тот же файл TypeScript содержит 681 токен на GPT-5.x и 1 178 токена на последнем токенайзере Claude — разница в 1,73 раза. Новый токенайзер Anthropic генерирует примерно на 30% бол","url":"https://www.aioga.com/ru/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:04.078Z"},"ar":{"title":"التكلفة الفعلية لنموذج الحدود: اختلافات التوكنيزير تؤدي إلى زيادات خفية في الأسعار","summary":"يحتوي نفس ملف TypeScript على 681 رمزا على GPT-5.x، و1,178 رمزا على أحدث رمز لكلود—أي فرق قدره 1.73 مرة. الرمز الجديد من Anthropic يولد حوالي 30٪ رموز أكثر من النسخة القديمة، مع ثبات في سعر السعر، مما يؤدي إلى زيادة خفية في السعر. كلود أوبوس 4.6 وأوبوس 4.8 كلاهما بسعر 5.00 دولار / 25.00 دولار، لكن الرمز الجديد يزيد عدد الرموز في نفس الكود بحوالي 32٪. سعر 2.00 / 10.00 دولار في كلود سونيت 5 هو سعر ترويجي، وبعد 31 أغسطس 2026، سيعود إلى 3.00 / 15.00 دولار، وعندها سيكون نفس الكود أعلى بنسبة 32٪ من سونيت 4.6. في المقارنات بين البائعين، يولد التوكنizer الجديد من كلود رموز أكثر من GPT بمقدار 1.50 إلى 1.73 ضعف الرموز في الكود، مع وجود أكبر فجوة في TypeScript.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"التكلفة الفعلية لنموذج الحدود: اختلافات التوكنيزير تؤدي إلى زيادات خفية في الأسعار - Aioga أخبار الذكاء الاصطناعي","description":"يحتوي نفس ملف TypeScript على 681 رمزا على GPT-5.x، و1,178 رمزا على أحدث رمز لكلود—أي فرق قدره 1.73 مرة. الرمز الجديد من Anthropic يولد حوالي 30٪ رموز أكثر من النسخة القديمة، مع ثبا","url":"https://www.aioga.com/ar/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:03.789Z"},"hi":{"title":"फ्रंटियर मॉडल वास्तविक लागत: टोकनाइज़र मतभेदों से छिपी हुई कीमत में वृद्धि होती है","summary":"उसी टाइपस्क्रिप्ट फ़ाइल में GPT-681 पर 5.x टोकन और क्लाउड के नवीनतम टोकनाइज़र पर 1,178 टोकन हैं - 1.73 गुना का अंतर। एंथ्रोपिक का नया टोकनाइज़र पुराने संस्करण की तुलना में लगभग 30% अधिक टोकन उत्पन्न करता है, जिसकी कीमत अपरिवर्तित रहती है, जिसके परिणामस्वरूप कीमत में वृद्धि होती है। क्लाउड ओपस 4.6 और ओपस 4.8 दोनों की कीमत $5.00/$25.00 है, लेकिन नया टोकनाइज़र एक ही कोड में टोकन की संख्या को लगभग 32% बढ़ा देता है। क्लाउड सॉनेट 5 का $2.00/$10.00 एक प्रचार मूल्य है, और 31 अगस्त, 2026 के बाद, यह $3.00/$15.00 पर वापस आ जाएगा, जिस बिंदु पर उसी कोड की कीमत सॉनेट 32 की तुलना में लगभग 4.6% अधिक होगी। क्रॉस-वेंडर तुलनाओं में, क्लाउड का नया टोकनाइज़र कोड में GPT की तुलना में 1.50x से 1.73x अधिक टोकन उत्पन्न करता है, जिसमें टाइपस्क्रिप्ट सबसे बड़ा अंतर दिखाता है।","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"फ्रंटियर मॉडल वास्तविक लागत: टोकनाइज़र मतभेदों से छिपी हुई कीमत में वृद्धि होती है - Aioga AI समाचार","description":"उसी टाइपस्क्रिप्ट फ़ाइल में GPT-681 पर 5.x टोकन और क्लाउड के नवीनतम टोकनाइज़र पर 1,178 टोकन हैं - 1.73 गुना का अंतर। एंथ्रोपिक का नया टोकनाइज़र पुराने संस्करण की तुलना में लगभग 30%","url":"https://www.aioga.com/hi/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:03.996Z"},"it":{"title":"Costo reale del modello frontier: le differenze tra i tokenizzatori portano ad aumenti nascosti dei prezzi","summary":"Lo stesso file TypeScript contiene 681 token su GPT-5.x e 1.178 token sull'ultimo tokenizzatore di Claude—una differenza di 1,73 volte. Il nuovo tokenizzatore di Anthropic genera circa il 30% di token in più rispetto alla versione precedente, con il prezzo invariato, causando un aumento nascosto del prezzo. Claude Opus 4.6 e Opus 4.8 sono entrambi prezzati a 5,00 $ / 25,00 $, ma il nuovo tokenizzatore aumenta il numero di token nello stesso codice di circa il 32%. Il prezzo promozionale di Claude Sonnet 5 è di 2,00 $ / 10,00 dollari, e dopo il 31 agosto 2026 tornerà a 3,00 $ / 15,00 dollari, momento in cui lo stesso codice costerà circa il 32% in più rispetto al Soneto 4,6. Nei confronti tra fornitori e fornitori, il nuovo tokenizzatore di Claude genera da 1,50 a 1,73 volte più token rispetto a GPT in codice, con TypeScript che mostra il gap più ampio.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Costo reale del modello frontier: le differenze tra i tokenizzatori portano ad aumenti nascosti dei prezzi - Aioga Notizie IA","description":"Lo stesso file TypeScript contiene 681 token su GPT-5.x e 1.178 token sull'ultimo tokenizzatore di Claude—una differenza di 1,73 volte. Il nuovo tokenizzatore di Anthropic genera c","url":"https://www.aioga.com/it/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:04.591Z"},"nl":{"title":"Frontiermodel werkelijke kosten: Verschillen in tokenizers leiden tot verborgen prijsstijgingen","summary":"Hetzelfde TypeScript-bestand bevat 681 tokens op GPT-5.x, en 1.178 tokens op Claude's nieuwste tokenizer—een verschil van 1,73 keer. De nieuwe tokenizer van Anthropic genereert ongeveer 30% meer tokens dan de oude versie, met de prijsprijs ongewijzigd, wat resulteert in een verborgen prijsverhoging. Claude Opus 4.6 en Opus 4.8 zijn beide geprijsd op $5.00 / $25.00, maar de nieuwe tokenizer verhoogt het aantal tokens in dezelfde code met ongeveer 32%. De $2,00 / $10,00 van Claude Sonnet 5 is een promotieprijs, en na 31 augustus 2026 zal deze terugkeren naar $3,00 / $15,00, waarna dezelfde code ongeveer 32% duurder is dan Sonnet 4.6. Bij cross-vendor vergelijkingen genereert Claude's nieuwe tokenizer 1,50 tot 1,73 keer meer tokens dan GPT in code, waarbij TypeScript het grootste verschil laat zien.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Frontiermodel werkelijke kosten: Verschillen in tokenizers leiden tot verborgen prijsstijgingen - Aioga AI-nieuws","description":"Hetzelfde TypeScript-bestand bevat 681 tokens op GPT-5.x, en 1.178 tokens op Claude's nieuwste tokenizer—een verschil van 1,73 keer. De nieuwe tokenizer van Anthropic genereert ong","url":"https://www.aioga.com/nl/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:02.810Z"},"tr":{"title":"Frontier modeli gerçek maliyeti: Tokenizer farklılıkları gizli fiyat artışlarına yol açar","summary":"Aynı TypeScript dosyası GPT-5.x'te 681 token ve Claude'un en son tokenizer'ında 1.178 token içeriyor—bu fark 1,73 kat. Anthropic'in yeni tokenizer'ı, eski sürüme göre yaklaşık %30 daha fazla token üretiyor, fiyat fiyatı değişmiyor ve bu da gizli bir fiyat artışına yol açıyor. Claude Opus 4.6 ve Opus 4.8 her ikisi de $5.00 / $25.00 fiyatlarında sunuluyor, ancak yeni tokenizer aynı koddaki token sayısını yaklaşık %32 artırıyor. Claude Sonnet 5'in $2.00 / $10.00 fiyatı promosyon fiyatıdır ve 31 Ağustos 2026'dan sonra $3.00 / $15.00 olarak geri dönecek; bu durumda aynı kod Sonnet 4.6'dan yaklaşık %32 daha yüksek olacak. Çapraz satıcı karşılaştırmalarında, Claude'un yeni tokenizer'ı kod olarak GPT'den 1,50 ila 1,73 kat daha fazla token üretiyor ve en büyük farkı TypeScript gösteriyor.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Frontier modeli gerçek maliyeti: Tokenizer farklılıkları gizli fiyat artışlarına yol açar - Aioga AI Haberleri","description":"Aynı TypeScript dosyası GPT-5.x'te 681 token ve Claude'un en son tokenizer'ında 1.178 token içeriyor—bu fark 1,73 kat. Anthropic'in yeni tokenizer'ı, eski sürüme göre yaklaşık %30 ","url":"https://www.aioga.com/tr/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:03.007Z"},"vi":{"title":"Chi phí thực tế của mô hình biên giới: Sự khác biệt của Tokenizer dẫn đến tăng giá ẩn","summary":"Cùng một tệp TypeScript chứa 681 mã thông báo trên GPT-5.x và 1.178 mã thông báo trên trình mã thông báo mới nhất của Claude—chênh lệch 1,73 lần. Tokenizer mới của Anthropic tạo ra nhiều token hơn khoảng 30% so với phiên bản cũ, với giá không thay đổi, dẫn đến mức tăng giá ẩn. Claude Opus 4.6 và Opus 4.8 đều có giá 5,00 đô la / 25,00 đô la, nhưng trình mã hóa mới làm tăng số lượng mã thông báo trong cùng một mã lên khoảng 32%. 2,00 đô la / 10,00 đô la của Claude Sonnet 5 là giá khuyến mại và sau ngày 31 tháng 8 năm 2026, nó sẽ trở lại 3,00 đô la / 15,00 đô la, tại thời điểm đó, mã tương tự sẽ có giá cao hơn khoảng 32% so với Sonnet 4.6. Trong so sánh giữa các nhà cung cấp, tokenizer mới của Claude tạo ra nhiều token hơn 1,50 đến 1,73 lần so với GPT trong mã, với TypeScript cho thấy khoảng cách lớn nhất.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Chi phí thực tế của mô hình biên giới: Sự khác biệt của Tokenizer dẫn đến tăng giá ẩn - Tin tức AI Aioga","description":"Cùng một tệp TypeScript chứa 681 mã thông báo trên GPT-5.x và 1.178 mã thông báo trên trình mã thông báo mới nhất của Claude—chênh lệch 1,73 lần. Tokenizer mới của Anthropic tạo ra","url":"https://www.aioga.com/vi/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:02.668Z"},"id":{"title":"Biaya aktual model perbatasan: Perbedaan tokenizer menyebabkan kenaikan harga tersembunyi","summary":"File TypeScript yang sama berisi 681 token di GPT-5.x, dan 1.178 token di tokenizer terbaru Claude—selisih 1,73 kali. Tokenizer baru Anthropic menghasilkan sekitar 30% lebih banyak token daripada versi lama, dengan harga harga tidak berubah, menghasilkan kenaikan harga tersembunyi. Claude Opus 4.6 dan Opus 4.8 keduanya dihargai $5.00 / $25.00, tetapi tokenizer baru meningkatkan jumlah token dalam kode yang sama sekitar 32%. $2.00 / $10.00 Claude Sonnet 5 adalah harga promosi, dan setelah 31 Agustus 2026, itu akan kembali ke $3.00 / $15.00, di mana kode yang sama akan berharga sekitar 32% lebih tinggi dari Sonnet 4.6. Dalam perbandingan lintas vendor, tokenizer baru Claude menghasilkan 1,50x hingga 1,73x lebih banyak token daripada GPT dalam kode, dengan TypeScript menunjukkan kesenjangan terbesar.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Biaya aktual model perbatasan: Perbedaan tokenizer menyebabkan kenaikan harga tersembunyi - Berita AI Aioga","description":"File TypeScript yang sama berisi 681 token di GPT-5.x, dan 1.178 token di tokenizer terbaru Claude—selisih 1,73 kali. Tokenizer baru Anthropic menghasilkan sekitar 30% lebih banyak","url":"https://www.aioga.com/id/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:03.833Z"},"th":{"title":"ต้นทุนที่แท้จริงของโมเดล Frontier: ความแตกต่างของ Tokenizer นําไปสู่การเพิ่มขึ้นของราคาที่ซ่อนอยู่","summary":"ไฟล์ TypeScript เดียวกันมีโทเค็น 681 โทเค็นบน GPT-5.x และ 1,178 โทเค็นบน tokenizer ล่าสุดของ Claude ซึ่งต่างกัน 1.73 เท่า โทเค็นใหม่ของ Anthropic สร้างโทเค็นมากกว่าเวอร์ชันเก่าประมาณ 30% โดยราคาไม่เปลี่ยนแปลง ส่งผลให้ราคาเพิ่มขึ้นที่ซ่อนอยู่ Claude Opus 4.6 และ Opus 4.8 มีราคาอยู่ที่ $5.00 / $25.00 แต่โทเค็นใหม่จะเพิ่มจํานวนโทเค็นในรหัสเดียวกันประมาณ 32% Claude Sonnet 5 เป็นราคาโปรโมชั่น $2.00 / $10.00 และหลังจากวันที่ 31 สิงหาคม 2026 จะเปลี่ยนกลับเป็น $3.00 / $15.00 ซึ่งรหัสเดียวกันจะมีราคาสูงกว่า Sonnet 32 ประมาณ 4.6% ในการเปรียบเทียบข้ามผู้ขาย โทเค็นใหม่ของ Claude สร้างโทเค็นมากกว่า GPT ในโค้ด 1.50 ถึง 1.73 เท่า โดย TypeScript แสดงช่องว่างที่ใหญ่ที่สุด","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"ต้นทุนที่แท้จริงของโมเดล Frontier: ความแตกต่างของ Tokenizer นําไปสู่การเพิ่มขึ้นของราคาที่ซ่อนอยู่ - ข่าว AI Aioga","description":"ไฟล์ TypeScript เดียวกันมีโทเค็น 681 โทเค็นบน GPT-5.x และ 1,178 โทเค็นบน tokenizer ล่าสุดของ Claude ซึ่งต่างกัน 1.73 เท่า โทเค็นใหม่ของ Anthropic สร้างโทเค็นมากกว่าเวอร์ชันเก่าประม","url":"https://www.aioga.com/th/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:04.186Z"},"pl":{"title":"Rzeczywisty koszt modelu Frontier: Różnice w tokenizerach prowadzą do ukrytych podwyżek cen","summary":"Ten sam plik TypeScript zawiera 681 tokenów na GPT-5.x oraz 1 178 tokenów na najnowszym tokenizatorze Claude'a — różnica 1,73 razy. Nowy tokenizator Anthropic generuje około 30% więcej tokenów niż poprzednia wersja, przy niezmienionej cenie, co skutkuje ukrytą podwyżką ceny. Claude Opus 4.6 i Opus 4.8 są wycenione na 5,00 USD / 25,00 $, ale nowy tokenizer zwiększa liczbę tokenów w tym samym kodzie o około 32%. Cena 2,00 / 10,00 USD w Claude Sonnet 5 to cena promocyjna, a po 31 sierpnia 2026 roku cena powróci do 3,00 USD / 15,00 USD, po czym ten sam kod będzie kosztował około 32% więcej niż Sonnet 4.6. W porównaniu między dostawcami nowy tokenizer Claude generuje 1,50 do 1,73 razy więcej tokenów niż GPT w kodzie, przy czym TypeScript wykazuje największą różnicę.","category":"技巧观点","source":"Hacker News 热门（buzzing.cc 中文翻译）","pageTitle":"Rzeczywisty koszt modelu Frontier: Różnice w tokenizerach prowadzą do ukrytych podwyżek cen - Aioga Wiadomości AI","description":"Ten sam plik TypeScript zawiera 681 tokenów na GPT-5.x oraz 1 178 tokenów na najnowszym tokenizatorze Claude'a — różnica 1,73 razy. Nowy tokenizator Anthropic generuje około 30% wi","url":"https://www.aioga.com/pl/news/cmrjrg5aj0202bis4f9pzo8d3/","contentTranslated":true,"sourceHash":"60e0d5b6f005fb14","translatedAt":"2026-07-19T11:51:04.918Z"}}}}