Sep 30 · AI Daily

Write code faster, review changes, and keep parallel work moving.

🧵 Here is a roundup of all the developer updates from OpenAI DevDay:

OpenAI has introduced Ultrafast mode in Codex, claiming it is up to 8x faster than Astra Standard and 4x faster than Astra Fast to speed up code generation and iteration. Documentation is available at https://developers.openai.com/api/docs/guides/ultrafast-mode.

OpenAI has launched Codex cloud environments, running Codex tasks in reusable cloud setups with repositories, dependencies, scripts, and settings pre-configured to reduce setup time and speed up launches. Agents continue running after users close their laptops, allowing progress tracking and task adjustments from a phone or another computer. Documentation is available at https://learn.chatgpt.com/docs/cloud.

The OpenAI developer account announced a major version update for Codex CLI, bringing new capabilities such as voice chat, agent progress tracking, and parallel work management. The new interface features a full-screen design to improve readability, with all functions operating entirely within the terminal. Details can be found at https://learn.chatgpt.com/docs/codex/cli.

OpenAI introduced the ability to open pull requests in ChatGPT Work or the Codex desktop app, view diffs, and ask Codex questions about the changes. Users can also ask Codex for feedback before sharing their own modifications. Documentation is available at https://learn.chatgpt.com/docs/code-review?surface=app.

OpenAI has launched ChatGPT Plugin extensions for building interactive panels, file viewers, and sidebar homepages. Plugins receive clearer feedback during the review process, and with support for the proposed MCP Events specification, events from connected applications can trigger automations within ChatGPT. Documentation is available at https://developers.openai.com/plugins/build/extensions.

Following a meeting with executives from OpenAI, Anthropic, Meta, Google, and Nvidia, U.S. President Trump released a voluntary industry AI agreement. Companies agreed to hire independent auditors to evaluate whether AI systems comply with their design intentions and pledged to prevent AI tools from unintentionally invading other technology systems.

Arena announced that Anthropic's Claude Sonnet 5.5 will be available for a limited-time test starting September 30 at 8:00 AM PT. Once launched, users can select it from the Direct Mode dropdown menu. Claude Sonnet 5.5 (High) will be available in Direct Mode for 48 hours, after which it will remain in Battle and Agent Mode.

Yubo argues that a series of releases including claude tag, grok bot, muse, cue, and dots are general agents with capabilities in coding, work, and personal assistance, representing a new category of technology.

A lot of the hype around AI (from its proponents) enthusiastically frames it as a kind of destruction, or even total erasure: "AI just killed XYZ" — "XYZ is dead" (in reality, this is almost never accurate).

The general public's perception of AI is largely shaped by claims of this nature. Backlash is inevitable.

Have you ever tried telling Claude, "We have the ability to do anything, please be braver"

And have you ever tried saying that to yourself

Gary Marcus criticized the White House "super intelligence" agreement, which Trump described as a model of "self-regulation," arguing that its essence is a corporate promise to avoid regulation and silence the public. He questioned the practical meaning of the word "independent" in the agreement, pointing out that so-called independent parties could be contractors chosen by the signing major companies with whom they have existing business ties. He also mentioned that the "pacing" topic discussed warmly two weeks ago has disappeared, and Dario, Sam, and Elon did not push for it.

Trump signed an executive order requiring executive department policy websites, policy documents, and press releases to stop using "artificial intelligence" or "AI" and instead use "Super Intelligence." Government agencies are not required to modify past regulations and documents. Following a lunch with tech CEOs, Trump stated that the industry should rely on self-regulation, backed by oversight from the Department of Justice and the FBI, and added that companies will provide resource support to communities where data centers are located.

Want to know everything?

https://openai.com/index/devday-2026-recap

According to Bloomberg, OpenAI is negotiating with investors to complete a transitional funding round of at least $30 billion ahead of an IPO, valuing the company at approximately $1.4 trillion.

Tesla FSD (Supervised) has received approval in Croatia, making it the eighth European country to permit the feature, with a rollout expected to begin soon. The approval relies on a regulatory exemption framework established by the Netherlands Vehicle Authority (RDW) this spring. A pan-European Union vote has been postponed from October 6 to no earlier than December, and Tesla needs to secure a qualified majority of at least 15 countries. Sweden and France have expressed opposition, while approval discussions with Ireland and Italy are ongoing. FSD (Supervised) is currently active in 16 markets globally.

OpenAI has released GPT-6.1 Sol, stating that it performs strongly in coding, computer use, and cross-application workflows at a price lower than GPT-6 Astra. Referenced content notes that it is targeted at complex refactoring, deep codebase investigation, and long-running agents, with cached inputs receiving a 95% discount compared to standard input pricing. A documentation link is included at https://developers.openai.com/api/docs/models/gpt-6.1-sol.

According to Business Insider, OpenAI CEO Sam Altman outlined a three-step plan to achieve commercial dominance at the developer conference in San Francisco, consisting of models, building, and distribution. This includes providing optimal models, opening internal tools to developers (such as cloud Codex, and APIs geared toward agents and rapid decision-making), and ultimately building a marketplace platform connecting businesses. On the same day, OpenAI disclosed that ChatGPT weekly users have surpassed 1.2 billion, and introduced a committed spending mechanism and Sign in with ChatGPT for partner products.

Trump announced the "super intelligence" White House agreement on Truth Social, signed by executives from Google, OpenAI, Anthropic, Meta, xAI, and Nvidia. The agreement requires companies to implement four layers of safeguards: establishing internal controls for frontier model safety, equipping internal teams, hiring external auditors, and setting up board committees for oversight. Zuckerberg stated that the board will independently review audit reports. The agreement is described as having "moral binding force," enforced through self-discipline and peer review, with the document noting it could potentially convert into laws and regulations later.

The author summarized the announcements from OpenAI DevDay 2026: the personal agent product Dots has rolled out to ChatGPT Pro, Business Premium, and Enterprise users, supporting collaboration across more than 4,000 apps; the new model GPT-6.1 Sol was launched at roughly one-seventh the task cost of Astra; and a $500 subscription was introduced while the multiplier for the $200 Pro allowance was cut from 20x to 10x, with the $500 tier set at 25x.