Every published Top AI Stories item tagged with OpenAI Codex, newest first.
Twenty-five Fields Medal winners, among them Terence Tao, Peter Scholze and Maryna Viazovska, published a joint declaration arguing that the race to solve famous problems as a benchmark is damaging mathematics itself. Rushed announcements leave no time for a proper writeup or for citing the prior work of others, they write, raising severe attribution and plagiarism questions and breaking the human chain that turns a proof into understanding. The Clay Mathematics Institute separately acknowledged that Navier-Stokes has apparently been settled, noting that its own rules govern how credit gets assigned.
Developers can now create a cloud agent in a single call by naming a task, a model, the tools it may use and where it runs. OpenAI hosts and maintains the harness itself, handling context management, tool use and subagents, and will either run the agent in its own sandbox or on the customer's infrastructure. This concedes a real piece of ground: the hard part of shipping an agent has been the scaffolding around the model rather than the model, and OpenAI is now selling that scaffolding instead of leaving it to open-source frameworks. The API is in public beta.
Andreas Thom posted on Mastodon that his own exchanges with ChatGPT may have fed one of the ten mathematical results OpenAI announced in August, calling the company's silence about its training data dishonest. OpenAI's write-up of that result, on non-sofic groups, was criticized for omitting recent work by Thom and Gabor Kun, and was quietly amended afterwards. He is the second researcher in a week to raise this, following New York University's Tristan Buckmaster, whose objection ran alongside the Navier-Stokes announcement.
OpenAI published a proof that smooth three-dimensional fluid motion can break down in finite time, resolving a Millennium Prize question open for roughly 90 years. It credits an unreleased internal model it calls significantly more capable than GPT-6 Astra, run as about 10,000 coordinating agents. NYU's Tristan Buckmaster, who published a related blowup result a day earlier with Anthropic's Levent Alpöge, says the pair drafted inside OpenAI's Codex and got no answer when he asked whether the model had trained on those sessions. OpenAI says no specific user data was accessed, while conceding it cannot rule out that de-identified usage data helped. Neither proof has been refereed.
OpenAI published internal measurements of how far coding agents have taken over its own research. As of mid-August the research organization used 3.1 agent-workdays of effort for every workday of human labor, up from below one in June, with the median researcher spending more than $600 a day on inference at API prices. OpenAI says it has met its goal of an automated research intern by September and is aiming at a full automated AI researcher by March 2028. Agents still need steering: more than half of successful four-to-eight-hour tasks took at least one human intervention.
OpenAI is moving GPT-5.6 toward broad release, with reports pointing to a mid-July general availability after a government-gated preview limited to about 20 partner organizations. An OpenAI engineer indicated the top 'Sol Ultra' tier — a mode that coordinates subagents to tackle complex work — is coming to Codex, the company's AI coding agent. GPT-5.6 splits into three durable tiers, named Sol, Terra, and Luna, spanning flagship reasoning down to fast, low-cost inference.
A widely upvoted bug report says OpenAI's Codex coding assistant writes enormous volumes of low-value logs to a local database — one developer measured about 37 terabytes in 21 days, which works out to roughly 640 terabytes a year, enough to wear out a typical consumer solid-state drive inside a year. The cause is a verbose logging default that captures internal traffic most users never need. Related complaints stretch back across several versions of the Desktop and command-line apps, and the issue remains open and unfixed.
OpenAI shipped a Record and Replay feature for its Codex desktop app: walk the agent through a workflow once — say, filing an expense or uploading a video with its thumbnail and subtitles — and Codex saves it as a reusable, editable "skill" it can run on its own. It is the most consumer-facing step yet toward computer-use agents that automate by demonstration instead of by code. The feature needs a paid ChatGPT account with Computer Use enabled and is not yet available in the EU, UK, or Switzerland.
OpenAI agreed to buy Ona — the German startup formerly known as Gitpod — and fold its team into the Codex division. Ona moves AI coding agents off a developer's laptop and into persistent cloud sandboxes that keep running even after the workstation shuts down, letting agents grind on multi-day tasks without interruption. The deal sharpens OpenAI's enterprise-coding push against Anthropic's Claude Code as the two labs race for developer mindshare. Terms were not disclosed.
OpenAI and Oracle said Oracle Cloud Infrastructure customers will be able to apply their existing Oracle Universal Credits toward OpenAI's frontier models and the Codex coding agent, putting that access inside the purchasing workflow enterprises already use. Availability is expected in the coming weeks. It is the latest thread in the deep Stargate compute relationship between the two companies, and it gives OpenAI another enterprise distribution channel — a reminder that the IPO-bound lab is still busy widening the on-ramps that drive its revenue.
Separately, OpenAI is planning its biggest ChatGPT redesign yet, reshaping the app into a "superapp" that pushes users toward its Codex coding tools, image generation, and partner services like Canva and Booking.com, the Financial Times reports. The revamp, rolling out in the coming weeks, is part of a broader pivot toward higher-margin enterprise revenue ahead of a possible IPO. Business customers already drive about 40 percent of OpenAI's revenue, a share the company expects to reach half by year-end.
OpenAI launched six role-specific Codex plug-ins — covering data analytics, creative production, sales, product design, equity investing, and investment banking — each bundling the integrations, instructions, and context to approximate a specific job. A new "Sites" feature turns Codex output into hosted interactive webpages via partners like Figma and Replit. OpenAI says knowledge workers are now roughly 20 percent of Codex users and growing more than three times as fast as its developer base.
OpenAI's frontier models and its Codex coding agent are now generally available on Amazon Bedrock, graduating from the limited preview that opened in April. GPT-5.5 runs in AWS's US East region and GPT-5.4 in US East and US West, callable through the Responses API with pricing that matches OpenAI's first-party rates. The move lets AWS customers reach OpenAI models inside their existing security, logging, and compliance controls — a notable thaw between OpenAI and a cloud rival to its main backer, Microsoft.
OpenAI permanently handed product strategy to co-founder Greg Brockman, who confirmed plans to merge ChatGPT, Codex, and the developer API into a single platform aimed at "the agentic future." Codex engineer Thibault Sottiaux will run the combined core product across consumer, enterprise, and developer surfaces. The reorganization lands three days before Google I/O 2026 and is widely framed against a potential OpenAI initial public offering before year-end.
OpenAI made **Codex available inside the ChatGPT mobile app on Thursday in preview**, with rollout across iOS and Android on every subscription tier including Free. Mobile Codex is scoped to remote workflow management — review outputs, approve commands, change models, switch threads, kick off new tasks — rather than the autonomous background execution the desktop CLI added last month. The change makes Codex meaningfully more ambient: engineers reviewing diffs on a phone is a different work pattern than reviewing them on a laptop, and free-tier availability lowers the bar for adoption beyond OpenAI's ChatGPT-paying base.
AWS and OpenAI announced on April 28 that GPT-5.5, GPT-5.4, the Codex coding agent, and a new Amazon Bedrock Managed Agents capability are now in limited preview for enterprise customers via Bedrock — OpenAI's first major distribution outside the seven-year Microsoft Azure exclusivity. Customers can evaluate and deploy OpenAI models alongside Anthropic, Meta, Mistral, Cohere, and Amazon's own models in a single Bedrock console with unified security, governance, and cost controls. The launch follows last week's amended Microsoft-OpenAI agreement that ended exclusive cloud rights through 2032 and signals a multi-cloud distribution era for frontier models.