codex

According to IT Home citing The Information, OpenAI may release the next generation of Astra and Sol series models, as well as the Cyber model aimed at the cybersecurity field, as early as November 6 (Friday) local time. The report did not disclose specific model numbers, pricing, context window lengths, or benchmark scores, nor is it clear whether these models will be launched first through API, ChatGPT, Codex, or partner channels.Previously, users had discovered the "gpt-6.2-astra" label in Codex usage statistics. Users suspected to be participating in the gray testing claimed that the new model has improvements in physical effects and animation generation, but these feedbacks are mainly based on subjective experiences, and there is currently no public benchmark data to verify.

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."

After Cursor and Anthropic, OpenAI announced its self-developed Git backend Pando, which has now taken over the company's entire main code repository. Over the past year, Codex has enabled engineers to deploy a large number of Agents to write code simultaneously, resulting in an increase in code branches, pushes, merges, and pull requests. OpenAI's original system experienced frequent stalls and merge failures, with reliability even dropping below 90%.One engineer is responsible for the core architecture design, and multiple Codex Agent teams participated in the implementation, with Pando going into production in 10 weeks. This backend allows code pushes from different branches to be processed in parallel without queuing each other. When pulling code, data packets are no longer generated separately for each client but instead reuse existing Git data packets, distributing them directly through caching. After the transformation, the median latency of code pushes was reduced by nearly 80%, distributing nearly 1 PB of data daily, with a cache hit rate exceeding 99%, but different branches no longer guarantee to present the same state at the same moment.Cursor's Origin is still based on Git, primarily aimed at replacing GitHub as a code hosting service. Anthropic's Scion, on the other hand, has abandoned Git, placing commits, merges, and historical queries on the server side. OpenAI retains the Git client and protocol, only reconstructing the backend, so engineers and Agents do not need to change their original development processes significantly.

On November 1, 2026, multiple users posted on X stating that OpenAI's Codex usage data included a model label named GPT-6.2 Astra, while the model selector still displayed GPT-6 Astra. A usage chart covering October 26 to November 1 showed that gpt-6-astra accounted for 80%, and gpt-6.2-astra accounted for 19.8%. Another account posted a suspected JSON snippet of an API response, with the response.model field being gpt-6.2-astra. According to Quantum Bit, GPT-6.2 Astra has been undergoing a limited gray testing among some users, with access primarily restricted to Pro plan users at $500 per month. Some users reported that about 20% of requests were routed to 6.2, while others claimed all requests pointed to 6.2.On the same day, an account posted a screenshot claiming that hidden code on the Anthropic official website's Mythos page included Claude Mythos 5.5. The visible page displayed Claude Mythos 5.1, while the hidden header and logo alt text indicated 5.5. According to Quantum Bit, this text briefly appeared in the front-end code of the official website and was removed a few hours later. Anthropic has not confirmed this, and currently, apart from the screenshot, there are no archived copies to verify this change.Additionally, there were previous reports that Fable 5.5 had appeared in the Microsoft Azure and MLflow directories, with a context window of 1 million tokens and an output limit of 128K tokens. Verification showed that on October 13, 2026, a bot added the entry azure_ai/claude-fable-5-5 to the price list of the open-source tool LiteLLM, which was deleted the next day, with the deletion note stating "Azure did not list this model." These parameters are identical to those of Fable 5.1. As of now, neither OpenAI nor Anthropic has officially released the aforementioned models.

Tibo from the OpenAI Codex team posted on X that the team will reopen the $200 Pro subscription to new users tomorrow and adjust the usage calculation method for the plan. According to his calculations, the adjusted subscription's corresponding API expenditure will be equivalent to half of the old $200 Pro plan.Tibo stated that the team does not wish to compromise in other ways and promised not to restore the 5-hour usage limit, allowing users to use their weekly quota as needed. He also mentioned that subscriptions should allow users to accomplish more work and achieve higher quality with the same expenditure as model efficiency improves and API prices decrease. This week, the team has reduced the prices of GPT-6 Sol and GPT-6 Luna to 50% of their previous prices and intends to continue lowering prices and enhancing capabilities, rather than raising API prices and then offering discounts.Tibo also revealed that tomorrow, several content items that do not consume usage quotas will be added to the subscription, but specific items have not yet been disclosed. He stated that this advance notice of the adjustment in the calculation method is to inform users of this change before subsequent announcements related to the subscription.

On September 2, B.AI announced that its Responses API is now integrated with the DeepSeek series of models. Developers can use the B.AI API Key to call the DeepSeek series of models supported by the B.AI platform in the Codex client.Previously, B.AI had supported calling the GPT series of models through Codex. After configuring the API Key and Base URL, developers can switch between DeepSeek and GPT for scenarios such as code generation, understanding, debugging, refactoring, and development collaboration.

Tibo, the product head of OpenAI Codex, stated that the number of active users has reached 25 million. To celebrate this milestone, the usage limits for all paid subscriptions of ChatGPT Work and Codex have now been reset.

On August 26, B.AI announced that the API is fully compatible with the OpenAI Responses API. Developers only need to configure the Base URL using the B.AI API Key to directly call the GPT series models supported by B.AI in the Codex development environment, seamlessly integrating cutting-edge large model capabilities into their daily coding workflow. This means that whether it's code generation, logical understanding, intelligent debugging, architectural refactoring, or team development collaboration, developers can obtain real-time intelligent assistance from the GPT series models in the Codex environment, comprehensively covering various scenario needs from individual development to team collaboration. Experience it now: http://b.ai

On August 24, a16z's latest chart shows that the weekly active growth of OpenAI Codex in enterprises is extremely polarized. The growth rate for lawyers is as high as 108 times, followed by sales and recruitment at 41 times each, marketing at 26 times, and healthcare at 24 times, while engineers are at the bottom with only 5 times.Codex has evolved from code completion to a digital colleague capable of independently breaking down tasks and running for hours, with lawyers using it to handle vast amounts of long texts and contract comparisons. This marks the "programming" of knowledge work, with the AI explosion shifting from the tech circle to high-value white-collar positions.

On August 21, OpenAI responded to the differences in Codex usage limits, stating that it would not unilaterally change the relevant rules without communicating with the community first. An official investigation found that affected users often converted subscriptions into API traffic through sub2api and redistributed it, which would be flagged by the fraud prevention system.OpenAI pointed out that using subscriptions normally through Sign in With ChatGPT on official or OSS clients (such as Pi, OpenCode) poses no issues at all.

On July 29, Tibo, the product lead for OpenAI Codex, announced on X that all usage limits for ChatGPT Work and Codex users have been reset, and released a statement regarding the issue of rapid consumption of GPT-5.6 Sol usage.In the previous weeks, a large number of users reported that the consumption speed of Codex limits exceeded expectations. OpenAI emphasized that it has not reduced the usage quotas of any subscription plans. After investigation, several reasons were found: Sol is more inclined to work for longer periods, execute more tool calls, and coordinate complex workflows than previous models; it consumes more tokens at the same reasoning level; programmatic tool calls (code mode) allow Sol to execute tool calls in parallel, resulting in more responses and higher usage per round; the distribution of impact is extremely uneven, with median user efficiency being acceptable, but heavy users handling complex tasks consuming far more than expected.OpenAI stated that multiple improvements have been completed, and it is expected that usage endurance will be extended by about 18% in typical usage scenarios. Tomorrow, the five-hour limit that was paused during the investigation will be restored. OpenAI acknowledged that "capabilities and efficiency do not always improve synchronously, and some issues only become apparent after large-scale real-world use; we should have realized this earlier."

According to the official status page of OpenAI, the widespread issues that ChatGPT and Codex experienced have been resolved, with the error rate remaining elevated for over 1 hour.

According to social media reports, OpenAI's ChatGPT and Codex are experiencing widespread outages.

OpenAI's programming agent Codex and office agent ChatGPT Work have a combined weekly active user count of 10 million. Just 9 days ago, it was 6 million. OpenAI subsequently reset the usage limits for paid users of both products.In March of this year, Codex had only 2 million weekly active users, and now the combined total for both products has expanded to 5 times.OpenAI had previously promised that for every additional 1 million users from 3 million to 10 million, the usage limits would be reset. This is the final milestone of this commitment.

On-chain analyst Specter issued a community alert on the X platform, indicating that the BNB Chain ecosystem project CodexField may have fraud and Rug Pull risks. The project has raised over $85 million with official support from BNB Chain.Specter tracked and found that the project wallet began transferring 17.3 million USDT from Tron to Ethereum, then exchanged it for DAI through Bitget Swap and transferred it to Polygon. Currently, $6.5 million has been transferred, while $10.8 million is still pending. The project operates two domain names and four subdomains for user deposits, and Blocksec MetaSuites has labeled the deposit contract as "Fake CodexField," but on-chain evidence shows that the contract is controlled by the project team. The flow of project funds is abnormal, as they are deposited into centralized exchanges through multiple chains and intermediary wallets, with just two Tron wallets having deposited over $3.4 million into exchanges. Specter calls on Binance founder Zhao Changpeng and BNB Chain to pay attention to this matter and possibly initiate an investigation.

According to official news, OpenAI has officially launched the preview version of the next-generation GPT-5.6 series models, including the flagship model Sol, the balanced model Terra, and the fast low-cost model Luna. GPT-5.6 introduces a brand new maximum reasoning effort and features a super strong mode that accelerates complex tasks through sub-agents.The flagship model Sol introduces the Ultra mode, which combines maximum reasoning intensity with sub-agent collaboration. In the Terminal-Bench 2.1 command line workflow test, Sol achieved a score of 88.8%, which increased to 91.9% in Ultra mode, surpassing GPT-5.5's 83.4% and Claude Fable 5's 88.0%. The mid-range model Terra performs close to GPT-5.5 while being priced at half, and the lightest model Luna is designed specifically for everyday automation tasks. Sol is priced at $5 per million input tokens and $30 for output, and it supports reducing secondary call costs by utilizing prompt caching.In terms of security, the security assessment confirmed that Sol did not exceed the critical thresholds of the Preparedness Framework cybersecurity. OpenAI has invested over 700,000 A100 equivalent GPU hours in automated red team exercises, equipping the entire series of models with a defense stack that includes rejection mechanisms, real-time abuse classifiers, and account-level audits. Although the current limited release follows the U.S. government's security framework, OpenAI emphasizes that it does not want a government-led access mechanism to become the long-term default model, as it would limit defenders' access to cutting-edge tools.