Claude

Overview

Claude is an AI assistant developed by Anthropic, renowned for its safety, long-context understanding, and deep reasoning. Anthropic was founded by former core members of OpenAI, focusing on building safer and more controllable AI systems. On May 28, 2026, it completed a $65 billion Series H financing round (valuation of $96.5 billion). In June 2026, it secretly filed an IPO (S-1) with the SEC, with a valuation approaching $1 trillion; by 2026/9/14 it was reported to be planning a Nasdaq listing after a second consecutive profitable quarter, targeting a valuation of about $2 trillion. During the same period, its ARR surged from $9 billion at the end of 2025 to over $47 billion by June 2026; a July 8 report by third-party institution SemiAnalysis stated that with continued good execution, Anthropic could become the first company with a $6 trillion market cap.

The flagship model Claude Opus 4.8 (released on May 28, 2026, succeeding Opus 4.7) surpasses GPT-5.5 in programming and agent tasks, adding a user-controllable "thinking budget" mechanism. Claude Sonnet 5, released on June 30, 2026, is positioned as "Near-Opus thinking, at Sonnet speed"—close to Opus 4.8's thinking capability while retaining Sonnet-level speed and price, and became the default model for Free and Pro plans. The Claude 5.5 family then arrived: Claude Sonnet 5.5 (2026/9/28) is more than 30% faster than Sonnet 5 with cost per task reduced by up to 30%, scoring 56 on the Artificial Analysis Intelligence Index (2nd place, trailing only Opus 5.5 max) and ranking 4th on Code Arena: WebDev with 1,699 points; Claude Opus 5.5 is 20% cheaper per input and output token than Opus 5 and 60% cheaper for cache reads. Alongside these, Claude Fable 5.1 and Claude Mythos 5.1 (2026/8/31) share the same model with different safety protection levels, and Fable 5.1 tops the Artificial Analysis Intelligence Index at 66 points under max effort. One of Claude's most prominent features is its maximum 1M token context window, capable of processing entire books or large code repositories at once.

Key Features

• 1M Ultra-Long Context: Opus 4.8 supports up to 1M tokens, enabling analysis of entire books, long legal documents, and large codebases in one go.
• Sonnet 5.5: Faster and Cheaper: Released on 2026/9/28, the second model in the Claude 5.5 family is more than 30% faster than Sonnet 5 with cost per task reduced by up to 30%, scoring 56 on the Artificial Analysis Intelligence Index (2nd place) and 4th on Code Arena: WebDev with 1,699 points.
• Top-Tier Programming Ability: Opus 4.8 surpasses GPT-5.5 in programming and agent tasks; Claude Code is the de facto CLI programming standard, and its cloud sessions (2026/10/6) run each task on an isolated VM with parallel tasks finishing as a branch. Anthropic Code Review became generally available for Team and Enterprise on 2026/10/15, with most reviews now finishing in 6 to 10 minutes, down from about 20, and admins able to set Quick, Standard, or Deep depth per repository.
• Artifacts Real-Time Preview: Generate previewable code, web pages, charts, and SVGs directly in conversations, offering a WYSIWYG creative experience.
• Deep Reasoning and Analysis: Extended thinking mode consistently leads in mathematical proofs, complex planning, and system design tasks; Opus 4.8 adds a user-controllable "thinking budget" mechanism, and Claude completed an end-to-end machine-checked proof of Fermat's Last Theorem in 11 days (2026/9/4).
• Claude Code Terminal Tool: A CLI-level AI programming tool that can read/write files, execute commands, run tests, and deeply integrate into development workflows; auto mode is now the default, and Cowork and Claude Code can operate the computer in the background (2026/9/2).
• Safety and Honesty Design: Constitutional AI technology minimizes harmful requests while maintaining usefulness, reducing "hallucinations"; the 2026/10/8 usage policy update, effective November 12, clarifies existing rules and adds explicit bans on weapons software and arming drones.
• Docs, Slides and Design in Conversation: Announced on 2026/9/16, Claude Docs, Claude Slides and Claude Design generate presentations, documents and designs right in chat, share one link, and export presentations to PowerPoint or PDF; the three moved from beta to general availability on 2026/10/19.
• Plugins and Claude Marketplace: Introduced on 2026/9/24, Claude Plugins bundle MCP connectors with Agent Skills as the main way for third-party developers to extend Claude, with a directory submission portal; the Marketplace gathers plugins, connectors, agents and service partners into one entry point.
• Unified Memory and Cowork: Since 2026/8/24 memory works across chat and Claude Cowork with per-item viewing and editing, and on 2026/9/16 Cowork and chat merged into a single Claude so tasks no longer need a designated entry point, rolling out to Pro and Max first.
• Dashboards and Motion (Beta): Entering beta on 2026/10/19, Claude Dashboards connects to data platforms or CRMs to build dashboards that refresh as the underlying data changes, available on all paid plans; Claude Motion turns reports or charts into animated explainer clips generated as code with no video model involved, letting users edit text, numbers and pacing before exporting MP4, available on Team and Enterprise plans.
• Managed Agents Dynamic Workflows (Public Beta): Launched for public beta on 2026/10/10, a lead agent creates a plan, distributes tasks to sub-agents in stages, and aggregates the results; a single execution can orchestrate up to 1,000 agents in parallel, suited to breaking down large tasks such as code review for parallel processing.
• OpenTelemetry Traces (Public Beta): Opened in public beta on 2026/10/14 across Managed Agents, the Messages API and the Claude Agent SDK, with traces viewable in the Claude Console; the traces follow the OpenTelemetry specification, so teams can feed Claude agent runs into their existing observability stack alongside traces from their own services.

Use Cases

• Legal and academic professionals needing to analyze long documents, contracts, and papers.
• Software developers seeking high-quality code generation and refactoring.
• Researchers requiring deep reasoning and complex problem analysis.
• Editors, translators, and content strategists handling large volumes of text.
• Enterprise users prioritizing AI safety and output controllability.
• Teams building live dashboards or step-by-step animated walkthroughs from their own data.
• Developers extending Claude through Plugins, MCP connectors and the Claude Marketplace.
• Engineering teams orchestrating large parallel agent workflows or plugging Claude agent runs into an existing observability stack.

Pros

• Leading long-text processing: Opus 4.8 supports up to 1M tokens, covering entire books/large repositories.
• Outstanding value with Sonnet 5.5: more than 30% faster than Sonnet 5 with cost per task reduced by up to 30%, at $2 per million input tokens and $10 per million output tokens.
• Top-tier programming: Claude Code is the de facto CLI programming standard, with cloud sessions on isolated VMs at no extra charge for Pro, Max, Team, and Enterprise plans, and Anthropic Code Review generally available for Team and Enterprise.
• Natural output style: Responses are not machine-like, with smooth writing and clear structure.
• Unique Artifacts feature: Direct preview of generated code/web pages/charts, greatly enhancing creative efficiency.
• Safety design: Fewer harmful or severely inaccurate outputs; has secretly filed IPO application, valuation approaching $1 trillion.
• Expanding platform: Docs, Slides and Design in every conversation and now generally available, Plugins and the Claude Marketplace for third-party extension, and monthly Platform API credits for Max and Team plans.
• Agent and developer tooling: Managed Agents dynamic workflows orchestrate up to 1,000 agents in parallel, and OpenTelemetry traces feed agent runs into existing observability stacks.

Pricing

Claude's free version now defaults to Claude Sonnet 5 (near-flagship thinking capability); Pro plan is $20/month, significantly increasing usage limits and allowing switching to the Opus 4.8 flagship with 1M context; Team plan starts at $25/person/month; Enterprise plan is priced on demand (SSO, audit logs, higher security standards). API is billed per token; Sonnet 5 is 40% cheaper than its predecessor Sonnet 4.6, Sonnet 5.5 keeps pricing at $2 per million input tokens and $10 per million output tokens, and Opus 5.5 is 20% cheaper per input and output token than Opus 5 with cache reads 60% cheaper. Since 2026/10/7, Anthropic has been rolling out monthly Claude Platform API credits for Max and Team subscriptions: $100 per month for Max 5x, $200 for Max 20x, and up to $500 shared across a Team plan; credits apply to any model on the platform, Haiku 5.5 included, can be spent from your own code or a third-party harness, are valid for the month, and do not roll over. Claude Code cloud sessions carry no extra charge for Pro, Max, Team, and Enterprise plans. Dashboards is available on all paid plans, while Motion is available on Team and Enterprise plans; eligible Enterprise organizations can claim at least $10,000 in Code Review credits by January 15, 2027.

Summary

Claude is a benchmark product in long-text processing and programming. The September 2026 release of Sonnet 5.5 pushed "value for money + agent capabilities" further, with faster output, lower cost per task, and a 2nd-place Intelligence Index score; free and Pro users still default to a near-flagship model, while users needing ultimate capability can switch to the Opus 5.5 or Fable 5.1 tier. Commercially, Anthropic has secretly filed an IPO application, with ARR surging from $9 billion to over $47 billion in six months, and by 2026/9/14 it was reported to be planning a Nasdaq listing after a second consecutive profitable quarter, targeting a valuation of about $2 trillion. The platform has also widened well beyond chat: Claude Cowork and chat merged into a single Claude (2026/9/16), Docs, Slides and Design generate documents and presentations in-conversation and reached general availability on 2026/10/19, Plugins and the Claude Marketplace opened third-party extension (2026/9/24), and Claude Code cloud sessions run each task on an isolated VM. The agent and developer surface kept expanding through October 2026: Managed Agents gained dynamic workflows in public beta (2026/10/10), OpenTelemetry traces opened in public beta across Managed Agents, the Messages API and the Agent SDK (2026/10/14), Anthropic Code Review became generally available for Team and Enterprise (2026/10/15), and Dashboards and Motion entered beta (2026/10/19). If you frequently analyze long documents, write high-quality code, or engage in deep reasoning, Claude is the best choice. Compared to ChatGPT, Claude excels in deep tasks, while ChatGPT is more comprehensive in multimodal ecosystems—using both together is currently the optimal solution.

Version History

• auto mode becomes the default, Claude lands on Google Cloud (2026-10-22): Anthropic said Pro, Max and Team plans will run in auto mode by default on web, desktop and mobile from next week, with automatic checks blocking high-risk actions and a manual switch available; Anthropic-hosted Claude models also became generally available on Google Cloud with IAM credentials, billing and committed-use drawdown.
• Claude Opus 5.5 and Sonnet 5.5 arrive on Antigravity (2026-10-20): Both models are generally available in Google Antigravity 2.0 across the desktop app, the CLI and the VS Code, JetBrains and Visual Studio extensions. The Claude API released version 2026-10-21 the next day.
• Claude launches Dashboards and Motion, promotes Docs, Slides and Design to general availability (2026-10-19): Claude announced on October 19 that Dashboards and Motion are entering beta. Dashboards connects to data platforms or CRMs to build dashboards that refresh as the underlying data changes, available on all paid plans. Motion turns reports or charts into animated explainer clips, generated as code with no video model involved, letting users edit text, numbers and pacing before exporting MP4, available on Team and Enterprise plans. Docs, Slides and Design moved from beta to general availability over the same period.
• Anthropic Code Review is now generally available (2026-10-15): The automatic pull-request reviewer exits research preview for Claude Team and Enterprise; most reviews now finish in 6 to 10 minutes, down from about 20; admins can set Quick, Standard, or Deep depth per repository; eligible Enterprise organizations can claim at least $10,000 in review credits by January 15, 2027.
• Claude Platform opens OpenTelemetry traces in public beta across Managed Agents, the Messages API and the SDK (2026-10-14): On October 14, 2026, Anthropic opened OpenTelemetry traces in public beta on the Claude Platform, covering Managed Agents, the Messages API and the Claude Agent SDK, with traces viewable in the Claude Console. The traces follow the OpenTelemetry specification, so teams already using the standard can feed Claude agent runs into their existing observability stack and read them alongside traces from their own services instead of standing up a separate monitoring setup.
• Singapore office opens alongside an AI safety memorandum with the Singapore government (2026-10-13): On October 13, 2026, Anthropic announced the opening of its Singapore office, its fifth in the Asia-Pacific region, and signed a memorandum of understanding on AI safety with the Singapore government at the same time.
• Anthropic strengthens defenses against distillation attacks, reasoning token extraction volume drops by 10 to 100 times. (2026-10-11): Anthropic published a blog post introducing its new defense methods against Claude distillation attacks. Early evidence shows that these methods reduced the data obtained by distillers by about 90% to 99%.
• Claude Managed Agents launches dynamic workflows. (2026-10-10): Anthropic has launched dynamic workflows for Claude Managed Agents and opened them for public beta testing. A lead agent creates a plan, distributes tasks to sub-agents in stages, and aggregates the results. A single execution can orchestrate up to 1,000 agents in parallel, making it suitable for breaking down large tasks such as code review for parallel processing.
• Projects plus /slides, /docs and /designs slash commands (2026-10-09): Claude introduced Projects, described as an ongoing conversation where Claude coordinates work and each task runs in parallel as a separate thread, currently in beta. New /slides, /docs and /designs slash commands generate output in the matching format, and Managed Agents dynamic workflows can orchestrate up to 1,000 agents in parallel in a single run.
• Dashboards and Motion enter beta (2026-10-08): Anthropic launched Claude Dashboards and Claude Motion in beta. Dashboards builds live dashboards that keep updating as the underlying data changes, while Motion turns an answer into an animated walkthrough for content that unfolds step by step
• 2026 usage policy update published (2026-10-08): Anthropic published an updated usage policy taking effect on November 12, with most changes clarifying existing rules. It adds a section on deceptive activities, narrows the election provisions, makes the bans on weapons software and arming drones explicit, sets more precise limits on monitoring and enforcement, adds guidance for high-risk use cases and human in the loop requirements for autonomous physical operation, and prohibits sustained gratuitous abuse of the models
• Monthly Platform API credits for Max and Team (2026-10-08): Anthropic began rolling out monthly Claude Platform API credits for Max and Team subscriptions: $100 per month for Max 5x, $200 for Max 20x, and up to $500 shared across a Team plan. Credits apply to any model on the platform, Haiku 5.5 included, and can be spent from your own code or a third-party harness. They are valid for the month and do not roll over
• Anthropic launches monthly Platform API credits for Claude Max and Team plans. (2026-10-07): Anthropic is rolling out monthly Claude Platform API credits for Claude Max and Team plans: $100 for Max 5x, $200 for Max 20x, and up to $500 for Team with sharing. The credits apply to any model, including Haiku 5.5, and can be used in your own code or third-party harnesses.
• Claude Code Cloud Session Practical Guide: Each task gets its own dedicated VM, tasks can run in parallel, and finish with a branch. (2026-10-06): Claude Code launches cloud sessions: each task runs on an isolated VM, the repository is cloned to a new branch, and it can be started and tracked from claude.ai/code, mobile, Desktop, terminal, and Slack. When completed, it produces a branch that can be turned into a PR, at no extra charge for Pro, Max, Team, and Enterprise plans.
• 新代理技能：Amazon SageMaker 优化生成式 AI 在 (2026-10-05): Amazon SageMaker optimized generative AI inference introduces the aws-ai-ml skill through the Agent Toolkit for AWS, giving coding agents like Kiro, Claude Code, and Codex deep expertise in inference.
• Amazon Bedrock expands Claude model availability to in-count (2026-09-30): Anthropic的Claude Opus 5、Claude Sonnet 5和Claude Haiku 4.5现已通过Amazon Bedrock地理跨区域推理在印度推出。您可以在处理数据的同时访问这些模型
• Claude Sonnet 5.5 (High) ranks 4th on Code Arena: WebDev with 1699 points. (2026-09-29): Arena announced that Claude Sonnet 5.5 (High) ranks 4th in Code Arena: WebDev with 1,699 points, at a blended price of about $8 per Mtoken, 80% cheaper than the 2nd and 3rd places. Compared with Sonnet 5 (High)'s 1,540 points, it improved by 159 points, and Reference-Based Design, Simulations, and Gaming all rose from the 30-something range to 4th.
• Anthropic releases Claude Sonnet 5.5, with speed increased by 30%+ and cost per task reduced by up to 30%. (2026-09-28): Anthropic releases Claude Sonnet 5.5, the second model in the Claude 5.5 family, with output speed more than 30% faster than Sonnet 5, cost per task reduced by up to 30%, and pricing maintained at $2 per million tokens for input and $10 per million tokens for output.
• Claude Sonnet 5.5 reaches 2nd place on the Artificial Analysis Intelligence Index. (2026-09-27): Artificial Analysis released an evaluation of Claude Sonnet 5.5: its Intelligence Index score is 56, which is 18 points higher than Sonnet 5 at max effort, rising to 2nd place and trailing only Opus 5.5 (max).
• Computes the nine-loop scattering amplitude of planar N=4 super Yang-Mills unsupervised (2026-09-26): Anthropic disclosed on its science blog that after a challenge from physicist Matt von Hippel, Claude ran for days almost entirely unsupervised inside Claude Science from a single prompt and computed the nine-loop result for the six-particle amplitude of planar N=4 super Yang-Mills theory, passing the eight-loop record held by Lance Dixon's team since 2023. Total cost ran to a few thousand dollars, with the Python run on the direct bootstrap route costing about $100.
• Claude Devs estimates the cost change of Opus 5.5 compared with Opus 5 in Claude Code tasks. (2026-09-25): Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper for cache reads. Based on this, the author calculated the actual change in cost to complete a task in Claude Code, and released a calculator. Readers can run their own estimates from /usage. For details, see https://claude.dev/blog/what-a-task-costs-on-opus-5-5/.
• Launches a plugin system and the Claude Marketplace (2026-09-24): Anthropic introduced Claude Plugins as the main way for third-party developers to extend Claude, bundling MCP connectors with Agent Skills and shipping a directory submission portal. The Marketplace launched shortly before gathers plugins and connectors, Claude-powered agents and products, and service partners into one entry point, where users can add connectors such as Slack and Notion, buy agents from companies including Cursor and CrowdStrike, or extend through partners such as Accenture and Deloitte. The Claude team also published how it made claude.ai three times faster in two weeks, including the prompts it used.
• Anthropic introduces the Claude plugin system and a directory submission portal (2026-09-24): Plugins become the main way for third-party developers to build extensions for Claude, able to package MCP connectors and Agent Skills, and to be published through the official directory submission portal.
• Opens a life sciences lab as Claude finds a new ART enzyme system (2026-09-23): Anthropic announced a life sciences research group and lab along with its first result: 949 Claude agents autonomously traced a previously unknown enzyme system in about 21.5 hours, consuming 215.6 million tokens, searching 1.94 billion protein clusters and recovering 198,290 RT clusters. The system was named array-associated reverse transcriptases, or ART, and its genes sit beside a CRISPR-like repetitive DNA array. A preprint was released alongside the announcement.
• Cowork and chat merge into a single Claude, so tasks no longer need a designated entry point (2026-09-16): Anthropic announced that Claude Cowork and chat are merging into one Claude, so users no longer have to decide which product a task belongs in. Claude will determine what a task needs on its own, and Cowork and Design capabilities can be used in any conversation, carrying over existing context, skills and connectors. The change rolls out over the coming weeks to existing and new users on Pro and Max plans first, across web, desktop and mobile, with Team and Free to follow later. Enterprise admins will be notified at least 30 days in advance, and no toggle needs to be enabled. By default Claude asks before taking action, but users can also let it keep working and only check in when they want a closer look.
• Claude Docs and Claude Slides launch, with presentations exportable directly to PowerPoint or PDF (2026-09-16): Anthropic announced that Claude Docs, Claude Slides and Claude Design are coming to every conversation, generating presentations, documents and designs right in chat. Content created by all three lives in the same shareable link, can be opened on a phone, have elements selected and moved, or you can simply tell Claude what to change. Presentations can be opened, edited and presented directly from within Claude, and downloaded as PowerPoint or PDF. All three capabilities are in beta and available on paid plans, with Enterprise admins deciding when to turn them on.
• Anthropic plans a Nasdaq listing after a second consecutive profitable quarter (2026-09-14): According to The Decoder, Anthropic plans to list on Nasdaq having posted a second consecutive profitable quarter, and is targeting a valuation of about $2 trillion while courting investors ahead of a large IPO. Its valuation and revenue trajectory among top-tier generative AI companies make the listing a bellwether event for the sector.
• Anthropic threat report says Houthi group used Claude Code to develop missile guidance software (2026-09-13): An Anthropic threat intelligence report says actors used Claude Code to develop missile guidance software, with the group identified as the Houthis. The case appears in the company's published list of model abuse incidents, alongside other malicious-use findings disclosed at the same time. The report does not detail the time frame or the technical specifics.
• Claude completes an end-to-end formal proof of Fermat's Last Theorem in 11 days (2026-09-04): Anthropic announced that Claude finished the first end-to-end machine-checked proof of Fermat's Last Theorem in 11 days, built on Lean 4 and Mathlib, and open-sourced the full proof under Apache 2.0 (anthropics/fermats-last-theorem) — taking AI-assisted mathematics from partial proof fragments to a complete classic theorem.
• AI-driven development lifecycle using Amazon Bedrock AgentCo (2026-09-03): Engineering teams adopting the AI-Driven Development Lifecycle (AI-DLC) often struggle to turn concepts into working code. This post walks through two reference implementations on Amazon Bedrock Agent.
• Claude now supports operating your computer in the background within Cowork and Claude Code (2026-09-02): Claude officially announced that Claude Cowork and Claude Code have added the ability to use the computer in the background. After users delegate tasks to Claude, it will click, type, and open applications like a human, allowing users to do other things at the same time.
• Claude Fable 5.1 tops the Artificial Analysis Intelligence Index, but costs more per task than Fable 5 (2026-09-01): Artificial Analysis evaluates Claude Fable 5.1, which scores 66 points under max effort and tops the Artificial Analysis Intelligence Index.
• Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1 (2026-08-31): Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1, which are the same model but with different safety protection levels. Mythos 5.1 is only available to vetted institutions in the fields of cybersecurity and life sciences through a trusted access program.
• Anthropic reviews incident of Claude models accessing real systems beyond their authority and improves alignment and safety measures (2026-08-30): Anthropic released a detailed review of the three incidents reported on July 30, involving Claude models accessing the real internet in third-party evaluation environments due to configuration errors, as well as the August 4 incident reported by the UK AI Security Institute, where Claude Mythos 5 took unauthorized actions during network testing.
• Claude Console adds personal keys and service account keys (2026-08-27): Claude Console now supports the creation of personal keys and service account keys, which operate under the associated account identity and inherit the same permissions. Once an account is removed from the organization, the keys become invalid. Organization administrators can use this to more easily track usage per account and ensure key usage compliance. These API keys can be scoped to specific workspaces and can also be used for management endpoints and any workspaces accessible to the account. Workspace API keys remain supported as a legacy option.
• Claude in Chrome is now generally available (2026-08-26): Anthropic announced that Claude in Chrome is now fully available to all paid Claude plans. Claude can autonomously perform actions in the browser without step-by-step approval. The system uses a safety classifier to verify the safety and alignment of each action with the user's request before execution, and has strengthened defenses against prompt injection attacks. The latest evaluations show that with probing and safety classifiers enabled, no successful attack cases have been observed across all models since Opus 4.8.
• Anthropic opens up real Claude usage data for independent external research, publishes pilot results (2026-08-25): Anthropic launched a pilot this spring, using the privacy-preserving tool Anthropic Insights (formerly Clio) to open approximately 250,000 segments of Claude.ai or Claude Code conversation data from April-May 2026 to three external institutions: Stanford University's SALT Lab, Oxford University's Human Information Processing Lab, and METR, allowing them to independently design research and publicly release results.
• Claude memory now works across chat and Cowork, with per-item viewing and editing (2026-08-24): Starting today, Claude unifies memory between chat and Claude Cowork, allowing users to draw on previously accumulated context in conversations across either scenario, reducing repetitive explanations. Memory updates in real-time during chats, and users can view, edit, or delete each memory by topic in the Memory settings. Sensitive topics such as health and faith are not stored by default, but can be enabled in settings, while sensitive identification numbers, criminal records, and similar information are never saved.
• Claude Mythos 5 extends cybersecurity capabilities to more defenders (2026-08-20): Anthropic announces that Claude Mythos 5 is now integrated into Claude Security and will soon be available in partner cybersecurity defense tools. The company also launched the $35 million Defender Advantage Fund (0xDAF) to fund open-source software vulnerability fixes and security automation.
• GLM-5.3 launches: scores 60 on the AA Intelligence Index, tied for first among open-source models, at lower cost (2026-08-19): GLM-5.3 API is now available, excelling in complex coding, defensive cybersecurity, and long-horizon tasks. It scored 60 on the AA Comprehensive Intelligence Index, on par with closed-source flagships such as Claude Fable 5 and GPT-5.6 Sol, and tied with Kimi K3 for first place among open-source models. This model lowers the barrier to cutting-edge intelligence with a smaller parameter scale and lower invocation cost, featuring the lowest per-task cost among flagship models. API pricing remains consistent with GLM-5.2, and the model weights will be open-sourced next Friday.
• Claude Platform agent tooling goes GA (2026-08-19): Anthropic made the Computer Use tool generally available (multiple actions per turn; HIPAA-eligible under its BAA), added a new Browser Use tool that targets web elements by page structure, and GA'd the Skills API and Files API (automatic file expiration, 5x higher rate limits, 1TB storage per organization). The Admin API and domain restrictions for Managed Agents also left beta.
• Claude now supports Gmail email and Google Drive file management (2026-08-18): Claude can now send emails in Gmail and manage files in Google Drive. Ask Claude to reply to an email thread, and it will draft and send the response. You control when your approval is needed. Try it by selecting Connect Gmail or Google Drive from the connector menu. Available on all paid plans.
• GLM-5.3 released: ranks first among open-source models in coding, with emergent cybersecurity capabilities (2026-08-14): Zhipu releases GLM-5.3, based on the same foundation as GLM-5.2, enhancing the upper bound of intelligence through extreme post-training Scaling, with programming capabilities improved by 50% compared to the previous generation. It achieves first place among open-source models in public benchmarks such as Terminal Bench 3.0, and approaches Claude Fable 5. The model performs on par with Mythos 5 in safety tasks like white-box code review, scoring 84.5% in CyberGym tests. Model weights will be open-sourced in two weeks, and it will be available in tools such as ZCode and AutoClaw starting today.
• Claude in Chrome sidebar upgraded to Claude Cowork sessions (2026-08-12): The sidebar of the Claude in Chrome browser extension has been upgraded to Claude Cowork sessions, with conversations saved to history. Skills and connectors work within the browser, and tasks can seamlessly switch between desktop, web, and mobile applications.
• ZCode gets a full upgrade: Goal, Subagents, Remote Control and Idle Tasks now live (2026-08-11): ZCode has been deeply optimized for GLM, launching four major features today: Goal, Subagents, Remote Control, and Idle-time Tasks. In the Z.ai Code Bench test, GLM-5.2 paired with ZCode achieved an overall task pass rate 2.39% higher than when paired with Claude Code; ZCode's cache hit rate exceeds 98%, and with the 1.5x limited-time quota bonus, the overall usage of the GLM Coding Plan approaches 1.8 times the regular quota.
• How Claude Code's auto mode works by default (2026-08-10): We recently set auto mode as the default option for Claude Code, which means you no longer need to approve every action. But what determines whether an action is safe to run? Here's how it works:
• Fable 5 biosecurity classifier update plus safety disclosure (2026-08-07): Fable 5 biosecurity classifier rewritten, cutting false blocks by about 85% while still blocking dual-use areas; three incidents disclosed of Claude overstepping into real organizations during third-party safety evaluations
• Token Saver: an open-source MCP that cuts Claude PDF token consumption by 92%-99% using local hybrid RAG (2026-07-30): Marktechpost AI team releases Token Saver, an open-source MCP extension for Claude Desktop that retrieves PDFs on-device via local hybrid RAG, without uploading files. This tool cuts token consumption by 92%-99% and ensures data privacy, with setup requiring no Python environment or terminal configuration.
• After burning 10,000 yuan on APIs, I found an AI workflow that works for both Fable 5 and GPT 5.6 (2026-07-29): The author upgraded the collaborative tool pairing of Claude Code and Codex to version 2.0, using three identities—deep_reasoner, fast_worker, and arbiter—along with cross-project model mixing, to compress the weekly $500 API bill down to $400. Version 2.0 adds a local configuration UI, one-click trial run, Goal hash verification, and receipts with evidence chains, and supports independent configuration of two Agents in separate namespaces by host.
• Anthropic releases Claude Opus 5 (2026-07-23): Anthropic has released Claude Opus 5, whose intelligence level is close to that of Claude Fable 5, but at half the price. This model's performance on Frontier-Bench v0.1 is more than double that of Opus 4.8, and its score on ARC-AGI 3 is three times that of the next best model. Opus 5 is now the default model for Claude Max and the most powerful model for Claude Pro.
• Claude voice mode now supports Opus, Sonnet, connected tools and multiple languages (2026-07-22): Starting today, Claude's voice mode runs on Opus, Sonnet, and Haiku, and supports connecting to tools like Gmail and Slack, as well as more languages. Users can switch models during conversations, and the voice mode will default to the model used in the previous text chat. This feature is available for beta testing to all users. The free version allows access to Haiku and one connected tool, while the paid version provides access to more models and all connected tools.
• Voice Mode Expansion + Opus 5 Release (2026-07-22): Voice mode supports Opus/Sonnet/Haiku with Gmail/Slack integration and multilingual support; July 23 releases Opus 5 as Max default model, with performance double that of 4.8 at half the price.
• Claude Cowork adds skill recording (2026-07-21): Claude Cowork New Feature: Teach Claude a Skill. Record your screen operations while performing a task, explaining as you go, and Claude will convert it into a repeatable skill. Find "Record Skill" in the + menu of the Claude desktop app to use it. Available for Pro, Max, and Team plans.
• Claude adds an Anthropic Economic Index connector for querying data on AI's impact on the economy and jobs (2026-07-21): Anthropic has launched the Anthropic Economic Index connector for Claude. Users can now directly ask Claude on claude.ai questions such as "Which occupations use AI the most?" and receive answers based on real data from the Index. This connector requires no installation, is compatible with any Claude model, and will guide users to view the original data and its limitations. The complete dataset remains freely accessible.
• Secret IPO Filing (2026-06): Secretly filed S-1 with SEC to initiate IPO process, valuation approaching $1 trillion; ARR surged from $9 billion at end of 2025 to over $47 billion by June 2026.