Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT 5.5 Spud comparison is not fact checkable: the OpenAI sources provided are a citation formatting guide and a Deep Research guide, not a model specif...
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT 5.5 Spud comparison is not fact checkable: the OpenAI sources provided are a citation formatting guide and a Deep Research guide, not a model specific source for GPT 5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citatio
Key takeaways
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narro
Research answer
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citation formatting and requires clearly visible, clickable inline citations for Deep Research outputs, while the supplied Anthropic sources document Claude Opus 4.7 as a flagship reasoning model and show workflow primitives such as files, prompt caching, and extended thinking, but do not in these excerpts show equally explicit end-user citation requirements [1][7][4][2][3][8]. Key findings - Anthropic describes Claude Opus 4.7 as part of its latest generation and as its most capable model for complex reasoning and agentic coding, with a “step-change jump” over Claude Opus 4 [4]. - Anthropic also states that Opus 4.7 uses a new tokenizer compared with previous models [5]. - For workflow inputs, Anthropic’s Files API lets developers upload and manage files for Claude API use without re-uploading the content on every request [2]. - For reasoning workflow mechanics, Anthropic documents special behavior for “thinking blocks” when using extended thinking with prompt caching [3]. Anthropic’s Claude Code docs also say that adding the word “ultrathink” to a skill enables extended thinking in that skill [8]. - On evidence provenance, OpenAI’s citation-formatting guide is explicitly about preparing citable material and instructing the model to format citations effectively [1]. - OpenAI’s Deep Research guide is more explicit about user-facing provenance: when web results, or information contained in those results, are shown to end users, inline citations should be clearly visible and clickable in the user interface [7]. - Inference: the Anthropic materials show building blocks that could support traceable research workflows, but the supplied excerpts do not explain how files, thinking blocks, or extended thinking are linked to user-visible citations or a source audit trail [2][3][8]. - For “scratchpads” specifically, the closest match in the supplied Anthropic evidence is “thinking blocks” under extended thinking [3]. The supplied OpenAI evidence does not describe a scratchpad mechanism; it focuses on citation formatting and Deep Research citation display requirements [1][7]. - Therefore, any claim that Claude Opus 4.7 or GPT-5.5 Spud is categorically better on citations, scratchpads, or end-to-end traceability would go beyond the supplied evidence. Insufficient evidence [1][2][3][4][7][8]. Evidence notes - The strongest provenance evidence in this set comes from vendor-primary documentation that directly addresses citation behavior, especially OpenAI’s citation-formatting guide and Deep Research guidance [1][7]. - The strongest Anthropic evidence in this set is vendor-primary documentation on model positioning and workflow components, including the model intro, Files API, prompt caching, and extended-thinking skill behavior [4][2][3][8]. Limitations / uncertainty - The evidence provided here is limited to vendor documentation snippets rather than full pages, and several snippets are partial, so claims beyond the visible text would be unsafe [1][2][3][4][5][6][7][8]. - Because the supplied materials do not include a model-specific page, benchmark, or technical documentation for GPT-5.5 Spud, an apples-to-apples model comparison is not possible from this evidence alone [1][7]. - The supplied Anthropic excerpts document reasoning-related mechanisms, but they do not establish, in the visible text, whether internal thinking artifacts are exposed to end users or mapped back to sources [3][8]. Summary The supported conclusion is limited but clear: in the supplied evidence, OpenAI is more explicit about end-user evidence provenance through citation formatting guidance and a requirement for visible, clickable inline citations in Deep Research outputs [1][7]. Anthropic is more explicit here about Claude Opus 4.7’s capability positioning and about workflow mechanics such as files, prompt caching, and extended thinking, but the supplied excerpts do not show an equally explicit public provenance standard for citations, scratchpads, or traceability [4][2][3][8]. A definitive Claude Opus 4.7 vs GPT-5.5 Spud verdict on research provenance is therefore not supported. Insufficient evidence [1][7].
Supporting visuals
Anthropic unveils Claude Opus 4.7 model with new cyber safeguardsA woman in a purple blazer appears to be speaking during a news broadcast about the announcement of the test release of Anthropic's Claude Opus 4.7, with a related stock market chart on the right showing a significant increase in private market value.Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude MythosLenovo Yoga Slim 7i Aura EditionAla Chere Seetha Ramuni Chentaku Now Streaming Online: Where to Watch, Plot, Cast, and MoreWhen an Upgrade Finally Feels Worth It: Samsung Galaxy A57 5GAs A Preferred Source On GoogleMotorola Edge 60 FusionGettyImages 300x200Gadgets360 Twitter ShareAkash Duttaanthropic unveils claude opus 4 7 enhanced ai model for coding and autonomous tasksAIFOD | AI FOR DEVELOPING COUNTRIES FORUMAIFOD | AI FOR DEVELOPING COUNTRIES FORUMHeadshot of Katelyn ChedraouiHeadshot of Katelyn ChedraouiHeadshot of Katelyn ChedraouiAIFODcropped pink aifod logo no text 1
Research conversation
You
Research question
Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability
Studio Global AI40 sources
Cited answer
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citation formatting and requires clearly visible, clickable inline citations for Deep Research outputs, while the supplied Anthropic sources document Claude Opus 4.7 as a flagship reasoning model and show workflow primitives such as files, prompt caching, and extended thinking, but do not in these excerpts show equally explicit end-user citation requirements [1][7][4][2][3][8]. Key findings - Anthropic describes Claude Opus 4.7 as part of its latest generation and as its most capable model for complex reasoning and agentic coding, with a “step-change jump” over Claude Opus 4 [4]. - Anthropic also states that Opus 4.7 uses a new tokenizer compared with previous models [5]. - For workflow inputs, Anthropic’s Files API lets developers upload and manage files for Claude API use without re-uploading the content on every request [2]. - For reasoning workflow mechanics, Anthropic documents special behavior for “thinking blocks” when using extended thinking with prompt caching [3]. Anthropic’s Claude Code docs also say that adding the word “ultrathink” to a skill enables extended thinking in that skill [8]. - On evidence provenance, OpenAI’s citation-formatting guide is explicitly about preparing citable material and instructing the model to format citations effectively [1]. - OpenAI’s Deep Research guide is more explicit about user-facing provenance: when web results, or information contained in those results, are shown to end users, inline citations should be clearly visible and clickable in the user interface [7]. - Inference: the Anthropic materials show building blocks that could support traceable research workflows, but the supplied excerpts do not explain how files, thinking blocks, or extended thinking are linked to user-visible citations or a source audit trail [2][3][8]. - For “scratchpads” specifically, the closest match in the supplied Anthropic evidence is “thinking blocks” under extended thinking [3]. The supplied OpenAI evidence does not describe a scratchpad mechanism; it focuses on citation formatting and Deep Research citation display requirements [1][7]. - Therefore, any claim that Claude Opus 4.7 or GPT-5.5 Spud is categorically better on citations, scratchpads, or end-to-end traceability would go beyond the supplied evidence. Insufficient evidence [1][2][3][4][7][8]. Evidence notes - The strongest provenance evidence in this set comes from vendor-primary documentation that directly addresses citation behavior, especially OpenAI’s citation-formatting guide and Deep Research guidance [1][7]. - The strongest Anthropic evidence in this set is vendor-primary documentation on model positioning and workflow components, including the model intro, Files API, prompt caching, and extended-thinking skill behavior [4][2][3][8]. Limitations / uncertainty - The evidence provided here is limited to vendor documentation snippets rather than full pages, and several snippets are partial, so claims beyond the visible text would be unsafe [1][2][3][4][5][6][7][8]. - Because the supplied materials do not include a model-specific page, benchmark, or technical documentation for GPT-5.5 Spud, an apples-to-apples model comparison is not possible from this evidence alone [1][7]. - The supplied Anthropic excerpts document reasoning-related mechanisms, but they do not establish, in the visible text, whether internal thinking artifacts are exposed to end users or mapped back to sources [3][8]. Summary The supported conclusion is limited but clear: in the supplied evidence, OpenAI is more explicit about end-user evidence provenance through citation formatting guidance and a requirement for visible, clickable inline citations in Deep Research outputs [1][7]. Anthropic is more explicit here about Claude Opus 4.7’s capability positioning and about workflow mechanics such as files, prompt caching, and extended thinking, but the supplied excerpts do not show an equally explicit public provenance standard for citations, scratchpads, or traceability [4][2][3][8]. A definitive Claude Opus 4.7 vs GPT-5.5 Spud verdict on research provenance is therefore not supported. Insufficient evidence [1][7].
People also ask
What is the short answer to "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability"?
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit
What are the key points to validate first?
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narro
Which related topic should I explore next?
Continue with "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Governance, auditability, and production controls for real deployments" for another angle and extra citations.
Cross-check this answer against "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Hallucination control, abstention behavior, and calibrated uncertainty".
Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. # Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. * Anthropic says the new AI model is less capable than Claude Mythos. * Anthropic’s First Cybersecurity AI Model Claude Mythos Is Here. In terms of internal benchmark evaluations, Anthropic claimed that the model performed better than OpenAI's GPT-5.4 and Google's Gemini 3.1 Pro. However, the scores reveal that the Opus 4.7 is st…
AI. * Claude. * Anthropic. Anthropic has announced its latest AI model with Claude Opus 4.7. Claude Opus 4.7 is the latest generally available version of Anthropic’s main AI model with a focus on advanced software development. However, Mythos isn’t generally available like Opus 4.7 since Anthropic is only sharing it with key software platform vendors like A…
Anthropic Unveils Claude Opus 4.7: Enhanced AI Model for Coding and Autonomous TasksAnthropic发布Claude Opus 4.7：增强型AI模型助力编程与自主任务Anthropic dévoile Claude Opus 4.7 : Modèle IA amélioré pour le codage et les tâches autonomesأنثروبيك تكشف عن كلود أوبوس 4.7: نموذج ذكاء اصطناعي محسّن للبرمجة والمهام الذاتية. Anthropic has announced the release of Claude Opus 4.7, the latest iteration of its advanced AI model tailored for complex coding and autonomous task execution. Anthropic a annoncé la sortie de Claude Opus 4.7, la dernière itération de son modèle IA avancé conçu pour l'exécution de tâches comp…
This copy is for your personal, non-commercial use only. Distribution and use of this material are governed by our Subscriber Agreement and by copyright law. For non-personal use or to order multiple copies, please contact Dow Jones Reprints at 1-800-843-0008 or visit www.djreprints.com. # Anthropic’s New Claude Update Brings Mythos Model Closer to Wider Release. Anthropic’s latest model, Claude Opus 4.7, is now generally available. Anthropic announced a new update to its Claude artificial intelligence model on Thursday, which the tech startup expects will help pave the way for the eventual d…
Anthropic's new Claude Opus 4.7 is its best public AI model — but not its most powerful. ## The company acknowledged the new model falls short of Claude Mythos Preview, which remains off-limits to the general public over cybersecurity concerns. Anthropic released Claude Opus 4.7, its most capable generally available AI model, while acknowledging the model is "less broadly capable" than Claude Mythos Preview, a more powerful system the company has declined to release publicly. During training, Anthropic took deliberate steps to pull back on what the model can do in cybersecurity contexts, a…
Claude Opus 4.7 正式发布：Anthropic 在推理模型上的又一次突破-51CTO.COM. # Claude Opus 4.7 正式发布：Anthropic 在推理模型上的又一次突破. 过去一周（4.10-4.17），AI 编程领域迎来了一波密集更新。Anthropic 发布 Claude Opus 4.7，OpenAI 给 Codex 应用添加了 computer use 和内置浏览功能，Cursor 推出 Canvases 交互式画布，Windsurf 2.0 直接集成了 Devin Cloud Agent。与此同时，各大 Code Agent（Claude Code、Cline、OpenCode、OpenClaw 等）也同步更新支持新模型。. #### Claude Opus 4.7 正式发布. Image 18: Claude Opus 4.7 发布公告Claude Opus 4.7 发布公告. Anthropic 于 4 月 16 日发布 Claude Opus 4.7，这是 Opus 系列的最新版本。根据官方说明，Opus 4.7 在编程、Agent、视觉和多步骤任务上都有更强表现，同时在长任务中的严谨性和一致性也有所提升。. Claude Code 也在 v2.1.111 中同步支持了 Opus 4.7 的 xhigh 模式，新增了`/effo…
雅虎香港財經. # Mythos不開放！Anthropic先推「較安全版」AI模型Claude Opus 4.7. 人工智慧公司 Anthropic 週四 (16 日) 宣布推出新一代模型 Claude Opus 4.7，強調在性能提升的同時，進一步降低高風險應用的潛在威脅。該公司指出，這款模型雖在整體能力上不及最新的 Claude Mythos Preview，但在軟體工程、指令遵循與實務任務完成等方面表現更佳，為目前最強大的可廣泛使用版本。. Anthropic 表示，Claude Opus 4.7 特別強化了代理式編程 (agentic coding)、跨領域推理、多工具運用及電腦操作等關鍵能力，在多項產業基準測試中均優於今年 2 月推出的前一代 Claude Opus 4.6。新模型同時維持與 4.6 相同的定價，並已全面整合至 Claude 產品線、應用程式介面(API)，以及透過微軟(MSFT-US)、Alphabet(GOOGL-US) 與亞馬遜 (AMZN-US) 等雲端平台提供服務。. 在安全設計方面，Anthropic 刻意降低 Claude Opus 4.7 的網路攻防能力，並導入自動偵測與阻擋機制，可即時攔截涉及違規或高風險資安用途的請求。公司表示，此舉旨在透過實際部署累積經驗，為未來更高階的 Mythos 級模型大規模釋出奠定基礎。. 相較之下，Cl…
| 聚焦進階軟體工程技術，Anthropic 釋出新模型 Claude Opus 4.7 |. 聚焦進階軟體工程技術，Anthropic 釋出新模型 Claude Opus 4.7. Anthropic 新款模型 Claude Opus 4.7 全面推出，並繼續將功能更強大的 Claude Mythos 限制給少數合作夥伴測試，用於網絡安全測試和修補軟體漏洞。. 相較上個版本 Claude Opus 4.6，最新 Claude Opus 4.7 在進階軟體工程方面有顯著進步，尤其在困難的任務上明顯提升。它以嚴謹且一致的方式處理複雜的長時間任務，精確注意指令，並在回報之前設計方法來驗證產出。用戶測試回報表示，那些以往需要密切監督、艱難的工作任務，可以放心交給它處理。. Claude Opus 4.7 的視覺能力也大幅提升，能以更高的解析度辨識圖像。而在處理專業任務時，它表現得更有品味和創意，能產出品質更高的介面、文件與簡報。. 雖然 Claude Opus 4.7 整體能力不及最強大的 Claude Mythos 預覽版，但下方的基準測試結果幾乎比 Claude Opus 4.6 更好。此外，比較 Claude Opus 4.7 與競爭對手 GPT-5.4，前者有占多數的 7 項測試結果領先後者。. ▲ Claude Opus 4.7 基準測試結果。. Anthropic…
Claude Opus 4.7 正式發布：程式碼、視覺雙升級，定價不變. Anthropic 於 2026 年 4 月 16 日發布 Claude Opus 4.7，SWE-bench Pro 拿下 64.3% 領先 GPT-5.4 與 Gemini 3.1 Pro，圖片解析度提升三倍，定價不變維持 $5/$25。但新版 tokenizer 最多讓實際成本增加 35%，升級前務必確認。. Claude Opus 4.7 正式發布：程式碼、視覺雙升級，定價不變. Anthropic 於 2026 年 4 月 16 日正式發布 Claude Opus 4.7，這是目前對外公開的最強 Claude 模型。程式碼能力與視覺解析度雙雙大幅提升，定價維持不變，但底層 tokenizer 異動讓實際使用成本有所不同，需要留意。. ## Opus 4.7 的核心升級是什麼. Claude Opus 4.7 是指 Anthropic 在 Opus 4.6 基礎上推出的新一代旗艦模型，model ID 為 i.j4i.i2
claude-opus-4-7
，主打三大方向：更強的程式碼解決能力、更高的視覺解析度，以及在長時間自主任務中更穩定的執行品質。. 程式碼能力是 Opus 4.7 改幅最大的面向。在業界標準測試 SWE-bench Verified（模擬解決真實 GitHub issue）上，Opus 4.…
This guide provides practical guidance on how to prepare citable material and instruct the model to format citations effectively, using patterns ... Mar 1, 2026
When displaying web results or information contained in web results to end users, inline citations should be made clearly visible and clickable in your user ...
The Deep Research API response includes a structured final answer along with inline citations, summaries of the reasoning steps, and source ... Jun 25, 2025
Deep research in ChatGPT uses reasoning to quickly gather, summarize, and interpret extensive information from across the web, helping you ... Apr 10, 2026
Affects citation accuracy and researcher reputation. Makes Deep Research unreliable for author-level fact-checking. Likely Cause: Deep Research ... May 2, 2025
Deep research is OpenAI's next agent that can do work for you independently—you give it a prompt, and ChatGPT will find, analyze, and synthesize ... Feb 2, 2025
All deep research outputs include citations or source links so you can verify the information. Completed research opens in a fullscreen report view designed ...
Here's a set of citations from a deep research article I generated. They all link to completely random websites that are entirely unrelated to the topic.
We analyzed millions of human-agent interactions across both Claude Code and our public API using our privacy-preserving tool, to ask: How much ... Feb 18, 2026
To study diffusion patterns, we introduce the Anthropic AI Usage Index (AUI) to measure whether Claude.ai use is over- or underrepresented in an ... 15 Sept 2025
At Anthropic, we view context engineering as the natural progression of prompt engineering. Prompt engineering refers to methods for writing and ... 29 Sept 2025
This data release aims to enable researchers and the public to better understand the economic implications of AI and investigate the ways in ... 15 Jan 2026
This is an early pilot risk report that reflects the kind of evidence and argumentation that we expect to need for future models that cross ... 26 Oct 2025
AI delegation approaches. Engineers and researchers are developing a variety of strategies for productively leveraging Claude in their workflow. 2 Dec 2025
Our Research feature involves an agent that plans a research process based on user queries, and then uses tools to create parallel agents that search for ... Jun 13, 2025
In this post, we share what we've learned from working with our customers and building agents ourselves, and give practical advice for developers on building ... 19 Dec 2024
Our system cards, which are published each time we release a model, provide analysis on some dimensions of risk—in particular, assessing our AI models for ...
This risk report argues that Claude Opus 4.6 does not pose a significant risk of autonomous actions that contribute significantly to later catastrophic ...
Avoid chain-of-thought prompts: Since these models perform reasoning internally, prompting them to “think step by step” or “explain your reasoning” is ...
Learn how to use OpenAI reasoning models in the Responses API, choose a reasoning effort, manage reasoning tokens, and keep reasoning state across turns.
This allows a reasoning summary that could be displayed to the end user to be returned at the same time as the raw CoT (which should not be ... Aug 5, 2025
In this work, we study whether current reasoning models are capable of controlling their chain of thought in ways that reduce monitorability. Mar 5, 2026
Instead, one should request a “reasoning summary” to obtain the desired value. ... How to handle the raw chain of thought in gpt-oss | OpenAI ... Aug 13, 2025
My goal is to use these reasoning texts and demonstrate them to our users in the Chat UI, similar to how ChatGPT shows its chain-of-thought. Jul 17, 2025
OpenAI introduces a new framework and evaluation suite for chain-of-thought monitorability, covering 13 evaluations across 24 environments. Dec 18, 2025
Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training ... Sep 12, 2024
Chain-of-thought (CoT) reasoning models “think” in natural language understandable by humans. Monitoring their “thinking” has allowed us to ... Mar 10, 2025
Chain-of-thought (CoT) monitoring is a promising tool for detecting misbehaviors and understanding the motivations of modern reasoning models. However, if.
The latest generation of Claude models: Claude Opus 4.7 - Our most capable model for complex reasoning and agentic coding, with a step-change jump over Claude ...
The token counting endpoint accepts the same structured list of inputs for creating a message, including support for system prompts, tools, images, and PDFs.
When using extended thinking with prompt caching, thinking blocks have special behavior: Automatic caching alongside other content: While thinking blocks cannot ...
The Claude Console also offers prompting tools—prompt generator, templates and variables, and prompt improver—to help you build and refine prompts quickly.
A content block that represents a file to be uploaded to the container Files uploaded via this block will be available in the container's input directory.
Converse API: Visual PDF analysis requires citations to be enabled. There is currently no option to use visual analysis without citations (unlike the ...
If you don't need to process text as it arrives, the SDKs provide a way to use streaming under the hood while returning the complete Message object, ...
If you are not using extended thinking, no changes are required. Thinking is off by default when you omit the thinking parameter. Prefer general instructions ...
The Claude model family ; Most capable. Opus 4.7 claude-opus-4-7. Best for complex analysis, coding, and creative tasks requiring deep reasoning. ; Best balance.
The techniques below are useful for general output consistency or when you need flexibility beyond strict JSON schemas. Here's how to make Claude's responses ...
To enable extended thinking in a skill, include the word “ultrathink” anywhere in your skill content. . Run skills in a subagent. Add context: fork to your ...
Opus 4.7 uses a new tokenizer compared to previous models, contributing to its improved performance on a wide range of tasks. This new tokenizer may use up to ...
If you're unsure which model to use, consider starting with Claude Opus 4.7 for the most complex tasks. It is our most capable generally available model, ...
We've launched the display field for extended thinking, letting you omit thinking content from responses for faster streaming. Set thinking.display: "omitted" ...
Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 have a 1M-token context window on Vertex AI. Other Claude models, including Sonnet 4.5 and Sonnet 4 ...
Our latest model, Claude Opus 4.7, is now generally available. Opus 4.7 shows improvements in software engineering and complex, long-running coding tasks, ...
The final result was a three-agent architecture—planner, generator, and evaluator—that produced rich full-stack applications over multi-hour ... Mar 24, 2026
With agent teams, multiple Claude instances work in parallel on a shared codebase without active human intervention. This approach dramatically ... Feb 5, 2026
This modular approach means you can update skills independently without rewriting agent logic, share Skills across multiple Agents, and scale capabilities as ...
In Claude 4 and later models, this limit applies to full thinking tokens, and not to the summarized output. However, when using interleaved thinking with tools, you can exceed this limit as the token limit becomes your entire context window. * Interleaved thinking amplifies cache invalidation, as thinking block…
Claude is capable of providing detailed citations when answering questions about documents, helping you track and verify information sources in responses. Provide document(s) and enable citations. By default, plain text and PDF documents are automatically chunked into sentences. If you need more control over citation granularity (e.g., for bullet points or transcripts), use custom content documents instead. For example, if you want Claude to be able to cite specific sentences from your RAG chunks, you should put each RAG chunk into a plain text document. Three document types are supported for…
These tools enable Claude to assist in hypothesis generation, experimental design, data analysis, and regulatory document automation .Early adopters (Biotech and Pharma companies) report substantial efficiency gains; for example, one large hospital system reduced oncology chart preparation time dramatically, and a major pharmaceutical partner cut clinical study report drafting from 12 weeks to 10 minutes using Claude-powered AI ( [10]www.beckershospitalreview.com ) ( [11]. Similarly, Baylor’s head of life sciences noted that patients effectively crowdsource ChatGPT with their data, giving it…
The first StrictlyVC of 2026 hits SF on April 30. # Anthropic’s new Citations feature aims to reduce AI errors. Citations is particularly useful in document summarization, Q&A, and customer support applications, Anthropic says, where the feature can nudge models to insert source citations. Citations isn’t available for all of Anthropic’s models — only Claude 3.5 Sonnet and Claude 3.5 Haiku. Anthropic notes that Citations may incur charges depending on the length and number of the source documents. StrictlyVC kicks off the year in SF. ### Blue Origin’s New Glenn put a customer satellite in the…
. With the new Claude 3.7 Sonnet, users can toggle “extended thinking mode” on or off, directing the model to think more deeply about trickier questions 1. That’s because we didn’t perform our standard character training on the model’s thought process. Both models start out in around the same place and get better with a larger number of steps…
This simple yet powerful technique—which, as we’ll explain below, is different from Claude’s new “extended thinking” capability (see here for extended thinking implementation details)—has resulted in remarkable improvements in Claude's agentic tool use ability. The reasoning Claude performs with the “think” tool is less comprehensive than what can be obtained with extended thinking, and is more focused on new information that the model discovers. The com…
In the first paper, we extend our prior work locating interpretable concepts ("features") inside a model to link those concepts together into computational "circuits", revealing parts of the pathway that transforms the words that go into Claude into the words that come out. As AI systems are rapidly becoming more capable and are deployed in increasingly important contexts, Anthropic is investing in a portfolio of approaches including [realtime monitoring](…
Extended Thinking is a mode where Claude internally analyzes a problem step by step before responding. In normal mode, Claude starts answering immediately. With
Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT 5.5 Spud comparison is not fact checkable: the OpenAI sources provided are a citation formatting guide and a Deep Research guide, not a model specif...
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT 5.5 Spud comparison is not fact checkable: the OpenAI sources provided are a citation formatting guide and a Deep Research guide, not a model specific source for GPT 5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citatio
Key takeaways
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narro
Research answer
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citation formatting and requires clearly visible, clickable inline citations for Deep Research outputs, while the supplied Anthropic sources document Claude Opus 4.7 as a flagship reasoning model and show workflow primitives such as files, prompt caching, and extended thinking, but do not in these excerpts show equally explicit end-user citation requirements [1][7][4][2][3][8]. Key findings - Anthropic describes Claude Opus 4.7 as part of its latest generation and as its most capable model for complex reasoning and agentic coding, with a “step-change jump” over Claude Opus 4 [4]. - Anthropic also states that Opus 4.7 uses a new tokenizer compared with previous models [5]. - For workflow inputs, Anthropic’s Files API lets developers upload and manage files for Claude API use without re-uploading the content on every request [2]. - For reasoning workflow mechanics, Anthropic documents special behavior for “thinking blocks” when using extended thinking with prompt caching [3]. Anthropic’s Claude Code docs also say that adding the word “ultrathink” to a skill enables extended thinking in that skill [8]. - On evidence provenance, OpenAI’s citation-formatting guide is explicitly about preparing citable material and instructing the model to format citations effectively [1]. - OpenAI’s Deep Research guide is more explicit about user-facing provenance: when web results, or information contained in those results, are shown to end users, inline citations should be clearly visible and clickable in the user interface [7]. - Inference: the Anthropic materials show building blocks that could support traceable research workflows, but the supplied excerpts do not explain how files, thinking blocks, or extended thinking are linked to user-visible citations or a source audit trail [2][3][8]. - For “scratchpads” specifically, the closest match in the supplied Anthropic evidence is “thinking blocks” under extended thinking [3]. The supplied OpenAI evidence does not describe a scratchpad mechanism; it focuses on citation formatting and Deep Research citation display requirements [1][7]. - Therefore, any claim that Claude Opus 4.7 or GPT-5.5 Spud is categorically better on citations, scratchpads, or end-to-end traceability would go beyond the supplied evidence. Insufficient evidence [1][2][3][4][7][8]. Evidence notes - The strongest provenance evidence in this set comes from vendor-primary documentation that directly addresses citation behavior, especially OpenAI’s citation-formatting guide and Deep Research guidance [1][7]. - The strongest Anthropic evidence in this set is vendor-primary documentation on model positioning and workflow components, including the model intro, Files API, prompt caching, and extended-thinking skill behavior [4][2][3][8]. Limitations / uncertainty - The evidence provided here is limited to vendor documentation snippets rather than full pages, and several snippets are partial, so claims beyond the visible text would be unsafe [1][2][3][4][5][6][7][8]. - Because the supplied materials do not include a model-specific page, benchmark, or technical documentation for GPT-5.5 Spud, an apples-to-apples model comparison is not possible from this evidence alone [1][7]. - The supplied Anthropic excerpts document reasoning-related mechanisms, but they do not establish, in the visible text, whether internal thinking artifacts are exposed to end users or mapped back to sources [3][8]. Summary The supported conclusion is limited but clear: in the supplied evidence, OpenAI is more explicit about end-user evidence provenance through citation formatting guidance and a requirement for visible, clickable inline citations in Deep Research outputs [1][7]. Anthropic is more explicit here about Claude Opus 4.7’s capability positioning and about workflow mechanics such as files, prompt caching, and extended thinking, but the supplied excerpts do not show an equally explicit public provenance standard for citations, scratchpads, or traceability [4][2][3][8]. A definitive Claude Opus 4.7 vs GPT-5.5 Spud verdict on research provenance is therefore not supported. Insufficient evidence [1][7].
Supporting visuals
Anthropic unveils Claude Opus 4.7 model with new cyber safeguardsA woman in a purple blazer appears to be speaking during a news broadcast about the announcement of the test release of Anthropic's Claude Opus 4.7, with a related stock market chart on the right showing a significant increase in private market value.Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude MythosLenovo Yoga Slim 7i Aura EditionAla Chere Seetha Ramuni Chentaku Now Streaming Online: Where to Watch, Plot, Cast, and MoreWhen an Upgrade Finally Feels Worth It: Samsung Galaxy A57 5GAs A Preferred Source On GoogleMotorola Edge 60 FusionGettyImages 300x200Gadgets360 Twitter ShareAkash Duttaanthropic unveils claude opus 4 7 enhanced ai model for coding and autonomous tasksAIFOD | AI FOR DEVELOPING COUNTRIES FORUMAIFOD | AI FOR DEVELOPING COUNTRIES FORUMHeadshot of Katelyn ChedraouiHeadshot of Katelyn ChedraouiHeadshot of Katelyn ChedraouiAIFODcropped pink aifod logo no text 1
Research conversation
You
Research question
Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability
Studio Global AI40 sources
Cited answer
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicitly documents citation formatting and requires clearly visible, clickable inline citations for Deep Research outputs, while the supplied Anthropic sources document Claude Opus 4.7 as a flagship reasoning model and show workflow primitives such as files, prompt caching, and extended thinking, but do not in these excerpts show equally explicit end-user citation requirements [1][7][4][2][3][8]. Key findings - Anthropic describes Claude Opus 4.7 as part of its latest generation and as its most capable model for complex reasoning and agentic coding, with a “step-change jump” over Claude Opus 4 [4]. - Anthropic also states that Opus 4.7 uses a new tokenizer compared with previous models [5]. - For workflow inputs, Anthropic’s Files API lets developers upload and manage files for Claude API use without re-uploading the content on every request [2]. - For reasoning workflow mechanics, Anthropic documents special behavior for “thinking blocks” when using extended thinking with prompt caching [3]. Anthropic’s Claude Code docs also say that adding the word “ultrathink” to a skill enables extended thinking in that skill [8]. - On evidence provenance, OpenAI’s citation-formatting guide is explicitly about preparing citable material and instructing the model to format citations effectively [1]. - OpenAI’s Deep Research guide is more explicit about user-facing provenance: when web results, or information contained in those results, are shown to end users, inline citations should be clearly visible and clickable in the user interface [7]. - Inference: the Anthropic materials show building blocks that could support traceable research workflows, but the supplied excerpts do not explain how files, thinking blocks, or extended thinking are linked to user-visible citations or a source audit trail [2][3][8]. - For “scratchpads” specifically, the closest match in the supplied Anthropic evidence is “thinking blocks” under extended thinking [3]. The supplied OpenAI evidence does not describe a scratchpad mechanism; it focuses on citation formatting and Deep Research citation display requirements [1][7]. - Therefore, any claim that Claude Opus 4.7 or GPT-5.5 Spud is categorically better on citations, scratchpads, or end-to-end traceability would go beyond the supplied evidence. Insufficient evidence [1][2][3][4][7][8]. Evidence notes - The strongest provenance evidence in this set comes from vendor-primary documentation that directly addresses citation behavior, especially OpenAI’s citation-formatting guide and Deep Research guidance [1][7]. - The strongest Anthropic evidence in this set is vendor-primary documentation on model positioning and workflow components, including the model intro, Files API, prompt caching, and extended-thinking skill behavior [4][2][3][8]. Limitations / uncertainty - The evidence provided here is limited to vendor documentation snippets rather than full pages, and several snippets are partial, so claims beyond the visible text would be unsafe [1][2][3][4][5][6][7][8]. - Because the supplied materials do not include a model-specific page, benchmark, or technical documentation for GPT-5.5 Spud, an apples-to-apples model comparison is not possible from this evidence alone [1][7]. - The supplied Anthropic excerpts document reasoning-related mechanisms, but they do not establish, in the visible text, whether internal thinking artifacts are exposed to end users or mapped back to sources [3][8]. Summary The supported conclusion is limited but clear: in the supplied evidence, OpenAI is more explicit about end-user evidence provenance through citation formatting guidance and a requirement for visible, clickable inline citations in Deep Research outputs [1][7]. Anthropic is more explicit here about Claude Opus 4.7’s capability positioning and about workflow mechanics such as files, prompt caching, and extended thinking, but the supplied excerpts do not show an equally explicit public provenance standard for citations, scratchpads, or traceability [4][2][3][8]. A definitive Claude Opus 4.7 vs GPT-5.5 Spud verdict on research provenance is therefore not supported. Insufficient evidence [1][7].
People also ask
What is the short answer to "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Evidence provenance in research workflows: citations, scratchpads, and traceability"?
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit
What are the key points to validate first?
Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narrower: OpenAI explicit Based only on the supplied evidence, a direct Claude Opus 4.7 vs GPT-5.5 Spud comparison is not fact-checkable: the OpenAI sources provided are a citation-formatting guide and a Deep Research guide, not a model-specific source for GPT-5.5 Spud [1][7]. What is supportable is narro
Which related topic should I explore next?
Continue with "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Governance, auditability, and production controls for real deployments" for another angle and extra citations.
Cross-check this answer against "Research and fact-check: Claude Opus 4.7 vs GPT-5.5 Spud, Hallucination control, abstention behavior, and calibrated uncertainty".
Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. # Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. Anthropic Releases Claude Opus 4.7 AI Model, Calls It Less Advanced Than Claude Mythos. * Anthropic says the new AI model is less capable than Claude Mythos. * Anthropic’s First Cybersecurity AI Model Claude Mythos Is Here. In terms of internal benchmark evaluations, Anthropic claimed that the model performed better than OpenAI's GPT-5.4 and Google's Gemini 3.1 Pro. However, the scores reveal that the Opus 4.7 is st…
AI. * Claude. * Anthropic. Anthropic has announced its latest AI model with Claude Opus 4.7. Claude Opus 4.7 is the latest generally available version of Anthropic’s main AI model with a focus on advanced software development. However, Mythos isn’t generally available like Opus 4.7 since Anthropic is only sharing it with key software platform vendors like A…
Anthropic Unveils Claude Opus 4.7: Enhanced AI Model for Coding and Autonomous TasksAnthropic发布Claude Opus 4.7：增强型AI模型助力编程与自主任务Anthropic dévoile Claude Opus 4.7 : Modèle IA amélioré pour le codage et les tâches autonomesأنثروبيك تكشف عن كلود أوبوس 4.7: نموذج ذكاء اصطناعي محسّن للبرمجة والمهام الذاتية. Anthropic has announced the release of Claude Opus 4.7, the latest iteration of its advanced AI model tailored for complex coding and autonomous task execution. Anthropic a annoncé la sortie de Claude Opus 4.7, la dernière itération de son modèle IA avancé conçu pour l'exécution de tâches comp…
This copy is for your personal, non-commercial use only. Distribution and use of this material are governed by our Subscriber Agreement and by copyright law. For non-personal use or to order multiple copies, please contact Dow Jones Reprints at 1-800-843-0008 or visit www.djreprints.com. # Anthropic’s New Claude Update Brings Mythos Model Closer to Wider Release. Anthropic’s latest model, Claude Opus 4.7, is now generally available. Anthropic announced a new update to its Claude artificial intelligence model on Thursday, which the tech startup expects will help pave the way for the eventual d…
Anthropic's new Claude Opus 4.7 is its best public AI model — but not its most powerful. ## The company acknowledged the new model falls short of Claude Mythos Preview, which remains off-limits to the general public over cybersecurity concerns. Anthropic released Claude Opus 4.7, its most capable generally available AI model, while acknowledging the model is "less broadly capable" than Claude Mythos Preview, a more powerful system the company has declined to release publicly. During training, Anthropic took deliberate steps to pull back on what the model can do in cybersecurity contexts, a…
Claude Opus 4.7 正式发布：Anthropic 在推理模型上的又一次突破-51CTO.COM. # Claude Opus 4.7 正式发布：Anthropic 在推理模型上的又一次突破. 过去一周（4.10-4.17），AI 编程领域迎来了一波密集更新。Anthropic 发布 Claude Opus 4.7，OpenAI 给 Codex 应用添加了 computer use 和内置浏览功能，Cursor 推出 Canvases 交互式画布，Windsurf 2.0 直接集成了 Devin Cloud Agent。与此同时，各大 Code Agent（Claude Code、Cline、OpenCode、OpenClaw 等）也同步更新支持新模型。. #### Claude Opus 4.7 正式发布. Image 18: Claude Opus 4.7 发布公告Claude Opus 4.7 发布公告. Anthropic 于 4 月 16 日发布 Claude Opus 4.7，这是 Opus 系列的最新版本。根据官方说明，Opus 4.7 在编程、Agent、视觉和多步骤任务上都有更强表现，同时在长任务中的严谨性和一致性也有所提升。. Claude Code 也在 v2.1.111 中同步支持了 Opus 4.7 的 xhigh 模式，新增了`/effo…
雅虎香港財經. # Mythos不開放！Anthropic先推「較安全版」AI模型Claude Opus 4.7. 人工智慧公司 Anthropic 週四 (16 日) 宣布推出新一代模型 Claude Opus 4.7，強調在性能提升的同時，進一步降低高風險應用的潛在威脅。該公司指出，這款模型雖在整體能力上不及最新的 Claude Mythos Preview，但在軟體工程、指令遵循與實務任務完成等方面表現更佳，為目前最強大的可廣泛使用版本。. Anthropic 表示，Claude Opus 4.7 特別強化了代理式編程 (agentic coding)、跨領域推理、多工具運用及電腦操作等關鍵能力，在多項產業基準測試中均優於今年 2 月推出的前一代 Claude Opus 4.6。新模型同時維持與 4.6 相同的定價，並已全面整合至 Claude 產品線、應用程式介面(API)，以及透過微軟(MSFT-US)、Alphabet(GOOGL-US) 與亞馬遜 (AMZN-US) 等雲端平台提供服務。. 在安全設計方面，Anthropic 刻意降低 Claude Opus 4.7 的網路攻防能力，並導入自動偵測與阻擋機制，可即時攔截涉及違規或高風險資安用途的請求。公司表示，此舉旨在透過實際部署累積經驗，為未來更高階的 Mythos 級模型大規模釋出奠定基礎。. 相較之下，Cl…
| 聚焦進階軟體工程技術，Anthropic 釋出新模型 Claude Opus 4.7 |. 聚焦進階軟體工程技術，Anthropic 釋出新模型 Claude Opus 4.7. Anthropic 新款模型 Claude Opus 4.7 全面推出，並繼續將功能更強大的 Claude Mythos 限制給少數合作夥伴測試，用於網絡安全測試和修補軟體漏洞。. 相較上個版本 Claude Opus 4.6，最新 Claude Opus 4.7 在進階軟體工程方面有顯著進步，尤其在困難的任務上明顯提升。它以嚴謹且一致的方式處理複雜的長時間任務，精確注意指令，並在回報之前設計方法來驗證產出。用戶測試回報表示，那些以往需要密切監督、艱難的工作任務，可以放心交給它處理。. Claude Opus 4.7 的視覺能力也大幅提升，能以更高的解析度辨識圖像。而在處理專業任務時，它表現得更有品味和創意，能產出品質更高的介面、文件與簡報。. 雖然 Claude Opus 4.7 整體能力不及最強大的 Claude Mythos 預覽版，但下方的基準測試結果幾乎比 Claude Opus 4.6 更好。此外，比較 Claude Opus 4.7 與競爭對手 GPT-5.4，前者有占多數的 7 項測試結果領先後者。. ▲ Claude Opus 4.7 基準測試結果。. Anthropic…
Claude Opus 4.7 正式發布：程式碼、視覺雙升級，定價不變. Anthropic 於 2026 年 4 月 16 日發布 Claude Opus 4.7，SWE-bench Pro 拿下 64.3% 領先 GPT-5.4 與 Gemini 3.1 Pro，圖片解析度提升三倍，定價不變維持 $5/$25。但新版 tokenizer 最多讓實際成本增加 35%，升級前務必確認。. Claude Opus 4.7 正式發布：程式碼、視覺雙升級，定價不變. Anthropic 於 2026 年 4 月 16 日正式發布 Claude Opus 4.7，這是目前對外公開的最強 Claude 模型。程式碼能力與視覺解析度雙雙大幅提升，定價維持不變，但底層 tokenizer 異動讓實際使用成本有所不同，需要留意。. ## Opus 4.7 的核心升級是什麼. Claude Opus 4.7 是指 Anthropic 在 Opus 4.6 基礎上推出的新一代旗艦模型，model ID 為 i.j4i.i2
claude-opus-4-7
，主打三大方向：更強的程式碼解決能力、更高的視覺解析度，以及在長時間自主任務中更穩定的執行品質。. 程式碼能力是 Opus 4.7 改幅最大的面向。在業界標準測試 SWE-bench Verified（模擬解決真實 GitHub issue）上，Opus 4.…
This guide provides practical guidance on how to prepare citable material and instruct the model to format citations effectively, using patterns ... Mar 1, 2026
When displaying web results or information contained in web results to end users, inline citations should be made clearly visible and clickable in your user ...
The Deep Research API response includes a structured final answer along with inline citations, summaries of the reasoning steps, and source ... Jun 25, 2025
Deep research in ChatGPT uses reasoning to quickly gather, summarize, and interpret extensive information from across the web, helping you ... Apr 10, 2026
Affects citation accuracy and researcher reputation. Makes Deep Research unreliable for author-level fact-checking. Likely Cause: Deep Research ... May 2, 2025
Deep research is OpenAI's next agent that can do work for you independently—you give it a prompt, and ChatGPT will find, analyze, and synthesize ... Feb 2, 2025
All deep research outputs include citations or source links so you can verify the information. Completed research opens in a fullscreen report view designed ...
Here's a set of citations from a deep research article I generated. They all link to completely random websites that are entirely unrelated to the topic.
We analyzed millions of human-agent interactions across both Claude Code and our public API using our privacy-preserving tool, to ask: How much ... Feb 18, 2026
To study diffusion patterns, we introduce the Anthropic AI Usage Index (AUI) to measure whether Claude.ai use is over- or underrepresented in an ... 15 Sept 2025
At Anthropic, we view context engineering as the natural progression of prompt engineering. Prompt engineering refers to methods for writing and ... 29 Sept 2025
This data release aims to enable researchers and the public to better understand the economic implications of AI and investigate the ways in ... 15 Jan 2026
This is an early pilot risk report that reflects the kind of evidence and argumentation that we expect to need for future models that cross ... 26 Oct 2025
AI delegation approaches. Engineers and researchers are developing a variety of strategies for productively leveraging Claude in their workflow. 2 Dec 2025
Our Research feature involves an agent that plans a research process based on user queries, and then uses tools to create parallel agents that search for ... Jun 13, 2025
In this post, we share what we've learned from working with our customers and building agents ourselves, and give practical advice for developers on building ... 19 Dec 2024
Our system cards, which are published each time we release a model, provide analysis on some dimensions of risk—in particular, assessing our AI models for ...
This risk report argues that Claude Opus 4.6 does not pose a significant risk of autonomous actions that contribute significantly to later catastrophic ...
Avoid chain-of-thought prompts: Since these models perform reasoning internally, prompting them to “think step by step” or “explain your reasoning” is ...
Learn how to use OpenAI reasoning models in the Responses API, choose a reasoning effort, manage reasoning tokens, and keep reasoning state across turns.
This allows a reasoning summary that could be displayed to the end user to be returned at the same time as the raw CoT (which should not be ... Aug 5, 2025
In this work, we study whether current reasoning models are capable of controlling their chain of thought in ways that reduce monitorability. Mar 5, 2026
Instead, one should request a “reasoning summary” to obtain the desired value. ... How to handle the raw chain of thought in gpt-oss | OpenAI ... Aug 13, 2025
My goal is to use these reasoning texts and demonstrate them to our users in the Chat UI, similar to how ChatGPT shows its chain-of-thought. Jul 17, 2025
OpenAI introduces a new framework and evaluation suite for chain-of-thought monitorability, covering 13 evaluations across 24 environments. Dec 18, 2025
Our large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training ... Sep 12, 2024
Chain-of-thought (CoT) reasoning models “think” in natural language understandable by humans. Monitoring their “thinking” has allowed us to ... Mar 10, 2025
Chain-of-thought (CoT) monitoring is a promising tool for detecting misbehaviors and understanding the motivations of modern reasoning models. However, if.
The latest generation of Claude models: Claude Opus 4.7 - Our most capable model for complex reasoning and agentic coding, with a step-change jump over Claude ...
The token counting endpoint accepts the same structured list of inputs for creating a message, including support for system prompts, tools, images, and PDFs.
When using extended thinking with prompt caching, thinking blocks have special behavior: Automatic caching alongside other content: While thinking blocks cannot ...
The Claude Console also offers prompting tools—prompt generator, templates and variables, and prompt improver—to help you build and refine prompts quickly.
A content block that represents a file to be uploaded to the container Files uploaded via this block will be available in the container's input directory.
Converse API: Visual PDF analysis requires citations to be enabled. There is currently no option to use visual analysis without citations (unlike the ...
If you don't need to process text as it arrives, the SDKs provide a way to use streaming under the hood while returning the complete Message object, ...
If you are not using extended thinking, no changes are required. Thinking is off by default when you omit the thinking parameter. Prefer general instructions ...
The Claude model family ; Most capable. Opus 4.7 claude-opus-4-7. Best for complex analysis, coding, and creative tasks requiring deep reasoning. ; Best balance.
The techniques below are useful for general output consistency or when you need flexibility beyond strict JSON schemas. Here's how to make Claude's responses ...
To enable extended thinking in a skill, include the word “ultrathink” anywhere in your skill content. . Run skills in a subagent. Add context: fork to your ...
Opus 4.7 uses a new tokenizer compared to previous models, contributing to its improved performance on a wide range of tasks. This new tokenizer may use up to ...
If you're unsure which model to use, consider starting with Claude Opus 4.7 for the most complex tasks. It is our most capable generally available model, ...
We've launched the display field for extended thinking, letting you omit thinking content from responses for faster streaming. Set thinking.display: "omitted" ...
Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 have a 1M-token context window on Vertex AI. Other Claude models, including Sonnet 4.5 and Sonnet 4 ...
Our latest model, Claude Opus 4.7, is now generally available. Opus 4.7 shows improvements in software engineering and complex, long-running coding tasks, ...
The final result was a three-agent architecture—planner, generator, and evaluator—that produced rich full-stack applications over multi-hour ... Mar 24, 2026
With agent teams, multiple Claude instances work in parallel on a shared codebase without active human intervention. This approach dramatically ... Feb 5, 2026
This modular approach means you can update skills independently without rewriting agent logic, share Skills across multiple Agents, and scale capabilities as ...
In Claude 4 and later models, this limit applies to full thinking tokens, and not to the summarized output. However, when using interleaved thinking with tools, you can exceed this limit as the token limit becomes your entire context window. * Interleaved thinking amplifies cache invalidation, as thinking block…
Claude is capable of providing detailed citations when answering questions about documents, helping you track and verify information sources in responses. Provide document(s) and enable citations. By default, plain text and PDF documents are automatically chunked into sentences. If you need more control over citation granularity (e.g., for bullet points or transcripts), use custom content documents instead. For example, if you want Claude to be able to cite specific sentences from your RAG chunks, you should put each RAG chunk into a plain text document. Three document types are supported for…
These tools enable Claude to assist in hypothesis generation, experimental design, data analysis, and regulatory document automation .Early adopters (Biotech and Pharma companies) report substantial efficiency gains; for example, one large hospital system reduced oncology chart preparation time dramatically, and a major pharmaceutical partner cut clinical study report drafting from 12 weeks to 10 minutes using Claude-powered AI ( [10]www.beckershospitalreview.com ) ( [11]. Similarly, Baylor’s head of life sciences noted that patients effectively crowdsource ChatGPT with their data, giving it…
The first StrictlyVC of 2026 hits SF on April 30. # Anthropic’s new Citations feature aims to reduce AI errors. Citations is particularly useful in document summarization, Q&A, and customer support applications, Anthropic says, where the feature can nudge models to insert source citations. Citations isn’t available for all of Anthropic’s models — only Claude 3.5 Sonnet and Claude 3.5 Haiku. Anthropic notes that Citations may incur charges depending on the length and number of the source documents. StrictlyVC kicks off the year in SF. ### Blue Origin’s New Glenn put a customer satellite in the…
. With the new Claude 3.7 Sonnet, users can toggle “extended thinking mode” on or off, directing the model to think more deeply about trickier questions 1. That’s because we didn’t perform our standard character training on the model’s thought process. Both models start out in around the same place and get better with a larger number of steps…
This simple yet powerful technique—which, as we’ll explain below, is different from Claude’s new “extended thinking” capability (see here for extended thinking implementation details)—has resulted in remarkable improvements in Claude's agentic tool use ability. The reasoning Claude performs with the “think” tool is less comprehensive than what can be obtained with extended thinking, and is more focused on new information that the model discovers. The com…
In the first paper, we extend our prior work locating interpretable concepts ("features") inside a model to link those concepts together into computational "circuits", revealing parts of the pathway that transforms the words that go into Claude into the words that come out. As AI systems are rapidly becoming more capable and are deployed in increasingly important contexts, Anthropic is investing in a portfolio of approaches including [realtime monitoring](…
Extended Thinking is a mode where Claude internally analyzes a problem step by step before responding. In normal mode, Claude starts answering immediately. With
. 
. Read more. Read more. Read more. [Rea…
. Developers can use 
. With the new Claude 3.7 Sonnet, users can toggle “extended thinking mode” on or off, directing the model to think more deeply about trickier questions 1. That’s because we didn’t perform our standard character training on the model’s thought process. Both models start out in around the same place and get better with a larger number of steps…