claude

The locally developed AI Agent Alma, led by independent developer Yetone, ranked third in the latest round of cross-evaluation initiated by the AI Agent infrastructure company Runta, both in task success rate and cost of success, outperforming tools like Claude Code and OpenCode. The current leaderboard snapshot shows that Alma has a task success rate of 72.7%, just behind Codex CLI at 77.9% and Pi at 72.8%, while Claude Code stands at 70.0%.Alma costs approximately $0.80 on average for each successfully completed task, compared to Claude Code's $0.94, making it about 16% cheaper. This FrontierHarness v1.1 comparison evaluates the performance of the same large model under different Agent execution frameworks, covering 21 sets of Harness configurations, 5 large models, and 62 tasks, with each task repeated 4 times, planning to run a total of 26,040 times, which is 72 times the scale of the previous version. All tools operate in a unified cloud environment, comparing task success rates, token costs, and execution speeds.Alma emphasizes local operation and long-term memory, capable of reading and modifying code, executing terminal commands, searching the web, and supporting Skills and MCP, while also allowing for switching underlying models. Recently, Alma integrated with Herdr, allowing programming Agents like Claude Code, Codex, and Pi to continue operating within Herdr, and enabling users to view execution processes, respond to questions, and approve operations in a centralized chat interface. Users can also remotely manage multiple Agents on their computers via iPhone. The evaluation is still ongoing, and rankings may change; the models being tested are limited to open-source models like GLM, DeepSeek, Kimi, and Muse, and do not cover closed-source models from the GPT and Claude series.

The new version of the "Supported Regional Policy" from the artificial intelligence company Anthropic will take effect on November 12. The new document specifies three categories of prohibited use of Claude, covering individuals, businesses, and users utilizing the model through overseas companies and third-party cloud platforms. As long as a person is in a non-supported region, they cannot use Claude; Americans using it during business trips in mainland China also violate the regulations. Even if a company is registered in the United States, if it is directly or indirectly controlled by individuals or businesses from non-supported regions, employees are similarly restricted, and it is not only those holding more than 50% of shares that will trigger this.The policy covers Claude.ai, Claude Code, Cowork, the official API, as well as Claude models provided by cloud platforms and authorized distributors. Calling the model through cloud platforms like AWS Bedrock does not mean that regional usage restrictions can be bypassed.As early as September 2025, Anthropic had already prohibited overseas companies controlled by businesses from non-supported regions like China from using Claude. This time, the policy clarifies the locations of individuals, the registration locations of businesses, and control relationships more explicitly. The official disclosure indicates that in the first half of 2026, a total of 11.4 million accounts were banned, with 398,000 appeals received, of which 42,000 successfully overturned the original decisions. The data covers various types of violations, not all of which involve regional restrictions. The system will identify risk signals from non-supported regions, unauthorized resale of Claude, and other issues, requiring relevant users to verify their identity; failure to verify will result in account suspension.

On November 10, AI company Anthropic disclosed in its IPO roadshow video that the company believes it can achieve an adjusted operating profit margin of 35% or more in the long term, up from 15% in the third quarter of 2026, but did not provide a specific timeline for achieving this. Chief Financial Officer Krishna Rao stated that the company plans to improve its long-term cost structure by enhancing computing efficiency, increasing the proportion of directly leased chips and data centers, and expanding the share of its inference business. The ratio of computing and infrastructure expenses to revenue aims to decrease from 58% in the third quarter to 45% to 50%; the proportion of sales, marketing, and partnership expenses is expected to drop from 14% to 7% to 10%, and adjusted personnel and other operating expenses are expected to decrease from 13% to 7% to 10%.As of the end of September, there were over 11,900 enterprise customers that contributed more than $100,000 in revenue to Anthropic over the past 12 months, including over 1,900 customers contributing more than $1 million and about 250 customers contributing over $10 million. The net revenue retention rate for enterprise customers is approximately 900%, with more than half of the enterprise customers using both the Claude Platform and at least one Anthropic proprietary application. The roadshow also showcased application cases from clients such as Eli Lilly, Rakuten, Salesforce, and Walmart. The company measures computing efficiency by the ratio of revenue to computing and infrastructure expenses, which increased from 0.40 times in the first quarter of 2025 to 1.72 times in the third quarter of 2026, approximately 4.3 times higher than before, but lower than the second quarter's 2.08 times. The company stated that it used about $5 billion in short-term computing in the third quarter, with costs significantly higher than those for computing procured under multi-year agreements.In terms of actual performance, Anthropic reported third-quarter revenue of $16.956 billion, adjusted operating profit of $2.463 billion, and an adjusted operating profit margin of 15%, down from 19% in the second quarter. Under GAAP standards, the company still recorded an operating loss of $567 million in the third quarter. The adjusted metrics exclude expenses such as equity incentives and non-cash charitable donations and cannot be equated to GAAP earnings. As of the end of October, the company's annualized revenue run rate was approximately $81 billion, not the actual revenue achieved for the full year. Additionally, Anthropic disclosed long-term commitments for computing and infrastructure of about $90 billion, of which approximately $45.16 billion is expected to occur by the end of 2030, with related commitments maturing in the next 12 months amounting to about $7.07 billion. As of the end of the third quarter, the company held approximately $96 billion in cash and cash equivalents and about $36 billion in short-term investments.

Elon Musk responded to user Dimdv's question on X, stating that the upcoming Grok 4.8 is "not good enough" compared to Anthropic's Claude Opus 5.5, and called "Opus 5.5 a masterpiece." He added that the next-generation model Grok 5 is expected to surpass Opus 5.5 in two to three months, but did not provide a comparative score or a specific release date.Musk also disclosed the progress of Grok 4.8: the model has completed pre-training and is still undergoing reinforcement learning, with the team spending significant time cleaning the supervised fine-tuning data. He mentioned that xAI has abandoned JAX and SGLang, opting for a reasoning and reinforcement learning system rewritten in C and C++, and stated that Starlink engineers are involved in building the relevant training and reasoning tech stack. Grok 4.8 is expected to be launched this week for a limited number of Bot users.Anthropic released Opus 5.5 on September 22, achieving performance close to its high-end model Fable 5.1 on most tasks, with operating costs under typical workloads being 40% lower than the previous generation Opus. In the Terminal-Bench 4 test, Anthropic reported that Opus 5.5 scored 66.4%, while OpenAI's GPT-6 Astra scored 57.9%.

AI company Anthropic has updated its "Supported Regions Policy" on its official website, with the new version taking effect on November 12. The policy applies to applications, APIs, and developer platforms such as Claude.ai, Claude Code, Cowork, as well as Claude models provided through cloud platforms and authorized distributors.The new policy clarifies that the following situations are considered as usage in unsupported regions: individuals located in unsupported regions, including users and employees of supported entities; entities registered or headquartered in unsupported regions and their users or employees, regardless of where these individuals are located; entities that are directly or indirectly majority-owned or controlled by individuals or entities from unsupported regions and their users or employees, again regardless of where they are located.According to the official website list, mainland China, Hong Kong, Macau, and Russia are not included in the list of supported regions.

On November 10, the B.AI platform announced the official launch of Anthropic Claude Fable 5.2, with API and Web Chat available simultaneously. This model is designed for complex reasoning, software engineering, and long-cycle agent workflows, always enabling adaptive thinking, equipped with a 1M Token ultra-large context window and a maximum output of 128K Tokens.According to the Artificial Analysis Intelligence Index v5 rating, Claude Fable 5.2 topped the list with a score of 65 under the highest configuration, far exceeding the median score of 40 among similar models. The generation speed has improved by up to 30% compared to Fable 5.1, with a standard price of $8.00 per million input Tokens and $40.00 per million output Tokens, and it supports prompt caching. You are welcome to log in to chat.b.ai/chat to experience it.

Renowned developer and T3 Code founder Theo discovered that although Anthropic's newly released Fable 5.2 is more powerful, the API equivalent usage available to subscription users is significantly lower. Theo has been tracking the actual request data of Claude Code for a long time. According to his estimates, for the same monthly subscription fee of $200 for Claude Ultra, using Opus 5.5 would yield an API usage equivalent to about $15,000 per month; switching to Fable 5.2 would only provide about $3,000, a difference of nearly 5 times.Anthropic stipulates that Ultra users can use a maximum of 50% of their total quota for Fable each week. Even if the entire quota is not used, as long as Fable reaches its limit, users must switch to other models or pay extra to continue using it. Theo estimates that if Opus 5.5 is used fully in a week, approximately 1,200 interaction rounds and 230 task sessions can be completed. If the Fable quota is exhausted first and then switched to Opus, the total number of interaction rounds would drop to about 900, although the number of completed task sessions might be slightly higher since Fable can sometimes complete tasks in one go without repeated communication.The API unit price of Fable 5.2 has decreased by 20% compared to the previous generation, with input and output tokens costing $8 and $40 per million, respectively, but it is still twice that of Opus 5.5. Anthropic reminds users that Fable will consume subscription quotas faster than other Claude models. Currently, Anthropic is offering eligible subscription users 1 to 3 opportunities to reset their Fable quota, with the number of resets depending on the subscription duration. Each reset can return the Fable quota consumed in the current week, allowing users to continue using it without waiting until the next week.

A federal judge in California, Alex Tse, ruled that the subscription marketing lawsuit against Anthropic's Claude Max can proceed. The judge allowed claims under California's consumer law for deceptive advertising, negligent misrepresentation, and breach of contract to continue, while dismissing other claims, including those related to the state's unfair competition law, and permitted the plaintiffs to submit an amended complaint.The proposed class action was filed in September by two Claude Max subscribers. The plaintiffs claim that the Max plans, priced at $100 and $200 per month, were advertised as providing 5 times or 20 times the usage of the Pro plan, but the weekly usage limits did not reach those multiples. Anthropic stated that the relevant multiples refer to the usage per 5-hour session and cited information disclosed on the website. Tse found that these disclosures were too far removed from the main sales page to warrant dismissal of the case at this stage.

Anthropic released Claude Fable 5.2 and the restricted version Mythos 5.2 on Monday, stating that the new model can transform a small number of instructions into runnable software, design physical objects, and address research questions that previously required a professional team. The official announcement claims that its output speed has improved by up to 30% compared to the previous generation, with Fable 5.2 scoring 49.1% in the internal FrontierCode 2 evaluation, higher than Fable 5.1's 32.4%. Both share the same underlying model, with Fable having stricter protections against dangerous biological and cybersecurity requests, while Mythos is only open to certified institutions in relevant fields.In terms of pricing, the standard API price is $8 per million input tokens and $40 per million output tokens, lower than the previous generation's $10 and $50; an optional fast mode can increase output speed by up to 2.5 times, but the rates double to $16 and $80. In comparison, Kimi 3.1 Pro is priced at $2 and $10, while the latest DeepSeek model is $0.6 and $1.98 during off-peak hours.AI blogger Matthew Berman's tests show that Fable 5.2 can generate playable games and a browser version of Excel. One level styled after "Gears of War" took 43 hours and 48 minutes to complete, consuming 8.63 billion tokens, with an estimated cost of $14,710 based on developer pay-per-use estimates; the Excel clone is estimated at about $11,700, and a "Fall Guys" style game ran for 71 hours, consuming over 9 billion tokens, approximately $13,000. Anthropic provides one week of Fable access for Pro users and standard Team seats, capped at half of the weekly quota, with promotional access for Pro users ending on November 16.

According to a report by the Korean News Agency, the American artificial intelligence company Anthropic has reported to the Personal Information Protection Commission of Korea that the account information of over 100 Korean enterprises for its generative AI service Claude has been leaked. The Commission stated on November 9 that Anthropic reported this matter on the 31st of last month, and it is currently in the preliminary review stage, with no formal investigation initiated yet.According to a previous report by The Washington Post, the American security startup Cathedral accessed Anthropic's internal systems by exploiting vulnerabilities in third-party office software. Anthropic researcher Lydia Hallie stated on X that Cathedral accessed account information such as names, emails, and packages without authorization, but did not obtain prompts, output results, or payment information. The company has blocked the affected systems and fixed the vulnerabilities on the day it became aware of the issue.

The Alibaba Cloud Token Plan Personal Edition will significantly increase the monthly quota. The prices for the Lite, Essential, and Standard tiers remain unchanged, while the quotas will increase by approximately 74% to 78%. The Lite package priced at 39 yuan will increase its quota from 11,500 to 20,000 Credits, the Essential package priced at 79 yuan will increase from 25,500 to 45,000, and the Standard package priced at 139 yuan will increase from 45,000 to 80,000.The Pro package will also see a price reduction and an increase in quota, with the monthly fee dropping from 499 yuan to 312 yuan, and the quota increasing from 180,000 to 200,000 Credits. Calculating the subscription cost per Credit, the costs for all four packages have decreased by approximately 43%.The Token Plan is Alibaba Cloud's monthly subscription service for large models, which can connect to AI programming and agent tools such as Claude Code, Qwen Code, Qoder, and OpenClaw. Users can call different models using a unified Credits quota. The new quotas will gradually be made available to existing users and will supplement the current subscription cycle's quota. The price for existing Pro users will remain unchanged for the current cycle, with the new price taking effect from the next cycle.

Anthropic launched the personal finance feature Personal Finance for Claude on Thursday, available in beta form for paid subscribers of Claude Pro, Max, and Ultra in the United States. Users can link their bank and credit card accounts through Plaid, which handles the login process, allowing Claude to access financial data without account passwords. This feature can summarize expenses, identify duplicate charges or unusual fees, explain interest expenditures, and generate charts or billing calendars. Anthropic stated that access is read-only, and Claude cannot transfer, manage, or invest user funds; Personal Finance does not include ads or sponsored recommendations.Cryptocurrency investor Anthony Pompliano raised allegations after the feature's release. His product Silvia is positioned as a personal AI Chief Financial Officer, and he claimed that Anthropic employees have been in contact with Silvia engineers since April, ostensibly to understand the company's growth and computing power needs, but in reality to probe into the product's construction and reasons for its popularity. He also described a steak dinner allegedly attended by representatives of Anthropic, claiming someone informed the Silvia team that the meal was paid for by CEO Dario Amodei. Pompliano stated that his engineers refused to disclose sensitive information and claimed that someone at Anthropic warned his team several days in advance about the upcoming product release, noting that Silvia's use of its own models and hardware incurs computing costs approximately 97% lower than cutting-edge AI models.Anthropic is not the first general chatbot company to enter the personal finance space. OpenAI launched a similar bank account linking feature for ChatGPT in May, also providing read-only access through Plaid. Anthropic stated that Personal Finance is just the first step and is exploring collaborations with licensed professionals and regulated financial institutions; currently, this beta version is limited to users of the three paid Claude subscription plans in the United States.

Former Anthropic Claude Code product lead Cat Wu posted on X that she will join OpenAI to manage core products. She stated that she respects OpenAI for demonstrating the effectiveness of a scalable path and bringing AI to over a billion people, believing that excellent products are at the heart of the mission.Cat Wu mentioned that she looks forward to collaborating with thsottiaux, the product management team, and colleagues at OpenAI to create agents that support individuals and teams while allowing users to maintain control. Her X profile has been updated to reflect her role in OpenAI's core products, with previous experiences including Anthropic's Claude Code, Dagster, and Scale AI. She previously posted that last Friday was her last day at Anthropic.

ByteDance recently introduced new internal management regulations for the use of AI tools, which will significantly restrict employees from accessing internal company data through external AI tools like Claude and Codex using personal accounts. It is reported that starting this month, ByteDance will implement a ban on external AI access for all L4 level code repositories and Feishu documents across the company, and AI accounts purchased by employees will no longer be able to access relevant internal resources.The notice was released internally at ByteDance last week, subsequently sparking discussions among some employees. The discussions mainly focused on the usage permissions of external AI tools, internal information security, and the relationship between the company's own AI products and third-party tools.ByteDance's relevant person in charge, Hong Dingkun, emphasized that internal information security is an important consideration for this adjustment, stating, "If L3 and above content is leaked, the consequences are very serious."