OpenAI's GPT-6.2 hits price parity
In ChatGPT, Codex, the API and GitHub Copilot
Nvidia rules out Jensen Huang's return. An Anthropic mathematician claims a proof of the Riemann hypothesis. Epic rebuilds Unreal Engine for AI coding. Claude solves two black hole problems. Microsoft builds its own coding models. And OpenAI launches GPT-6.2.

ON AIR
ACN · AI Coding News, the podcast. New every weekday.
Anthropic has closed its IPO, at eighty-three dollars a share.
Hackers published two malicious versions of Microsoft's Agent Framework on the Python package index.
OpenAI says its agents' estimated success rate on two- to four-hour tasks has doubled, to forty-six percent.
A cyberattack has disrupted the European Commission for more than ten days.
Google's Chrome team built a synthetic web to test AI agents safely.
An outage took down Microsoft's Outlook, Teams and Copilot for nearly two days.
Sponsor
Nvidia rules out Jensen Huang's return. An Anthropic mathematician claims a proof of the Riemann hypothesis. Epic rebuilds Unreal Engine for AI coding. Claude solves two black hole problems. Microsoft builds its own coding models. And OpenAI launches GPT-6.2. From AI Coding News, this is your weekly AI news recap for the week of November 16th.
Our top story: Nvidia's board has ruled out bringing back Jensen Huang as CEO. Huang resigned on Thursday, November 12th, after the disclosure of personal behavior inconsistent with Nvidia's code of conduct. He has since offered to return. On Tuesday, November 17th, chair Harvey Jones said on the earnings call that the board does not intend to reinstate him. On the same call, interim CEO Colette Kress ended Nvidia's sales of its H200 AI chips in China. Why it matters: the chipmaker at the center of AI now sets its course without its founder.
Next, a claim from mathematics. Anthropic mathematician Levent Alpöge says he has proved the Riemann hypothesis, working with Claude. The problem has stood open for one hundred sixty-seven years. On Friday, November 20th, he posted the proof in Lean, a programming language that checks each step. No one has verified it yet. Scientific American reports experts should be able to check it quickly. Why it matters: if the proof holds, an AI model helped crack one of math's best-known problems.
Next, Epic Games. Epic is rebuilding its Unreal Engine so AI coding agents can work inside it. A developer preview of Unreal Engine 5.9 arrived on Monday, November 9th, with Unreal Canvas, a space for working with AI agents. Unreal Engine 6 will drop its visual Blueprints for Verse, a text-based language. IEEE Spectrum reports Epic is racing standalone AI coding tools. One studio founder says those tools made leaving Unreal trivial. Why it matters: for studios on Unreal, the engine is being rebuilt so agents can do more of the coding.
Next, Anthropic. The company says Claude Fable 5.2 has solved two open problems in black hole physics. In August, an Anthropic researcher asked Claude for a list of open problems in the field. Claude came back with twenty-two, and its agents worked on them in the background for about six weeks. One result finishes a calculation physicists set up in 2003 and never completed. The other sets an exact limit on the mass of the lightest black hole, in three-dimensional gravity. A physicist checked every main result, and some were verified in Lean. Anthropic says neither settles quantum gravity, but each pushes physics forward. Why it matters: Claude picked the problems itself and solved two of them, one open for twenty-three years, largely on its own in six weeks.
Next, Microsoft. At its Ignite conference on Tuesday, November 17th, Microsoft launched a family of AI models of its own. Two are for coding, MAI Code 2 and MAI Code 2 Flash, and both are in GitHub Copilot. Microsoft says MAI Code 2 nearly matches OpenAI's GPT-6.1 Sol, at forty percent of the cost. A Greyhound Research analyst notes that comparison uses a half-price promotion. Why it matters: Copilot users get a cheaper coding model that Microsoft built itself.
When we come right back: OpenAI's new GPT-6.2 models. Your coding agents say they're done. Are they? Levr is the tracker and control plane that verifies and gates your coding agents' work against the plan. Levr holds the plans, issues, tests and gates for every agent doing the work. And work only moves forward when the gates are verified, not when the agent claims it's done. Levr. Try it free at levr.one. Welcome back.
And finally, OpenAI. On Tuesday, November 17th, OpenAI launched GPT-6.2 Astra, for its hardest work, and GPT-6.2 Sol, for everyday tasks. Both are in ChatGPT Work, Codex and the API. GitHub added them to Copilot the same day, for agentic coding and code review. Astra's price is twenty percent lower than GPT-6 Astra's. The Decoder reports the benchmark firm Artificial Analysis found Astra flat on its coding agent index. Why it matters: OpenAI's top model now costs less, and its coding agent score held flat.
Now, quick hits. Anthropic has closed its IPO, at eighty-three dollars a share. Hackers published two malicious versions of Microsoft's Agent Framework on the Python package index. OpenAI says its agents' estimated success rate on two- to four-hour tasks has doubled, to forty-six percent. A cyberattack has disrupted the European Commission for more than ten days. Google's Chrome team built a synthetic web to test AI agents safely. And an outage took down Microsoft's Outlook, Teams and Copilot for nearly two days.
The bottom line: AI agents took on more of the hard work themselves this week, in physics and maybe in math, and the models that write your code got cheaper. That's your brief for Friday, November 20th. Sources are linked in the description. ACN is brought to you by Levr, the tracker and control plane that holds your coding agents' work to the plan, at levr.one.