Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
OpenAI Releases Hundreds of Math Solutions Without Peer Review - OpenSmartRoute
OpenAI Releases Hundreds of Math Solutions Without Peer Review
OpenAI plans to dump hundreds of AI-solved math problems on GitHub without publishing papers. Mathematicians want formal verification and proper credit before accepting the results.
Key points
OpenAI claims to have solved over 100 long-standing open math problems.
The company plans to release these solutions on Tuesday via GitHub.
Mathematicians demand peer-reviewed papers instead of blog posts or tweets.
Tristan Buckmaster accused OpenAI of front-running his work on a Millennium Prize problem.
Why it matters: Engineers and managers must verify AI outputs before trusting them as facts or building systems on top of them.
By OpenSmartRoute editorial · written through the router by writer-small
From Wired AI - “OpenAI Is Pissing Off a Bunch of Mathematicians—Again”
OpenAI plans to release hundreds of math solutions on GitHub without peer review. The company will dump these results directly onto a public code repository. Mathematicians want formal papers and proper credit before accepting them. They fear this approach bypasses traditional scientific standards.
In August, OpenAI met with about 40 mathematicians to discuss the future. Attendees heard that powerful models solved hundreds of long-standing problems. Company representatives said they would not release everything at once. A spokesperson named Lindsay McCallum claimed the company was unaware of this plan. She worried how the community might react to such a move.
The group asked OpenAI to publish papers instead of blog posts or tweets. Bryna Kra from Northwestern University recalled the meeting as both exciting and dreadful. She noted that the firm ignored their request for formal publication. Mathematicians need time to absorb, digest, and use complex results properly.
OpenAI intends to post tens of thousands of mathematical solutions on GitHub this year. Frontier models are becoming increasingly capable at solving problems. The release will happen soon according to people familiar with the plans. This follows a pattern of direct model outputs rather than scientific papers.
On August 28, OpenAI began training a new internal model. The company claims this model resolved more than 100 open problems. It also solved the Navier-Stokes Millennium Prize problem according to McCallum. The firm is drawing advice from an advisory group at the Institute for Advanced Study. They have not set a specific release time yet.
Leading mathematicians feel the field has become a playground for tech companies. Anthropic and OpenAI are racing to show off their models before stock offerings. Traditional scientific processes for releasing results are being cast aside in this rush. Many believe these companies learn very little from past controversies.
In September, OpenAI deployed thousands of agents to solve the Millennium Prize problem. This happened after hearing rumors that others were closing in on solutions. Tristan Buckmaster accused the company of front-running his work toward solving an element. He had done this in a personal collaboration with Anthropic employee Levent Alpöge.
Buckmaster claimed OpenAI researcher Sébastien Bubeck implied Alpöge should be excluded from any paper. Bubeck denied asking for Alpöge to not be listed as an author when asked. Buckmaster refused to negotiate under these terms and threatened to tell the press.
Mistral launched Mistral Large 4, nicknamed Le Chonk. It is a 1 trillion-parameter model available for free.
Meeting notes show Bubeck worried about what Anthropic might do next. He suggested telling his leadership that OpenAI could get a Millennium Prize problem. There is no stopper for Anthropic giving all their compute to solve it too.
Bubeck also seemed to suggest Buckmaster would ruin his career if he spoke out. The notes record Bubeck saying he would not be nice unless Buckmaster was nice back. This dynamic creates a perception of mobster behavior among mathematicians according to Nestor Guillen.
OpenAI assembled an advisory group in mid-September to help determine how to communicate results. The goal is to uphold academic and professional standards while building research tools. Mathematicians say the company has made little progress on these goals since then.
Many are frustrated that OpenAI releases results through blog posts instead of scientific papers. This makes verification harder and often excludes prior work from other mathematicians. Piecemeal announcements are not limited to just the labs in San Francisco or Mountain View.
Levent Alpöge announced he disproved an 87-year-old conjecture through a tweet sent shortly after the World Cup final. He did not publish a formal paper for this specific achievement. Kra says math by tweet is not the way to nurture the ecosystem that created the data.
Some have set up new tools this year in reaction to the increase in machine-assisted proofs. Hexagon is a repository for primarily AI-generated material. Palomar is a registry of machine-verified mathematics. These tools help the community use new results and sort through what needs further digestion.
OpenAI has been directly encouraged to use these tools like Hexagon and Palomar. Kra notes they have not changed their behavior despite this encouragement. She feels the company's actions are not aligned with the Leiden declaration either.
The Leiden declaration calls for AI companies to meet mathematicians' standards. More than 4,000 mathematicians signed this call for better collaboration. It sets expectations for how AI should interact with mathematical research and learning.
Several employees within OpenAI believe their technology has rendered math dead anyway. McCallum says they do not believe the future of mathematics is set. They are working with the math community to navigate the future collaboratively.
The scenario where AI capabilities outpace human researchers was framed as hypothetical during the August meeting. Attendees who spoke to WIRED interpreted it as a warning about what was to come. The company briefed mathematicians about its capabilities in the same way police notify families before reporting a death.
Bubeck appears to believe his firm's technology will swiftly end most mathematicians' careers. He tells WIRED he believes more capable AI could help mathematicians tackle more ambitious questions. Better connections between work could solve real-world challenges according to him.
Mathematicians know they need to adapt, especially to equip younger generations. They do not believe their field is dead despite the threats. Kra says it changes how we operate but allows us to think bigger.
She is happy to have a new, powerful tool at her disposal to solve problems. She just wants companies' disclosures to be better so she can trust and build on the results. It is a scary time according to many in the field. But it is also really a deeply exciting time for those who adapt.
OpenAI Announces Plan to Release Math Solutions on GitHub
OpenAI plans to dump hundreds of math solutions on GitHub next Tuesday. People familiar with the plan tell WIRED about this release. The company will share results for unsolved problems online. This move follows earlier posts of ten problems in August. Attendees at the meeting wanted formal papers instead. They asked OpenAI to publish work properly.
OpenAI spokesperson Lindsay McCallum says the firm is not aware of all details. She told attendees they would not release everything at once. The company sought advice from experts before publishing. Mathematicians felt this assurance was ignored by leadership. The advisory group at the Institute for Advanced Study gave recommendations. OpenAI has not set a specific release time yet.
The internal model trained on August 28 started new work. It resolved the Navier-Stokes Millennium Prize problem according to McCallum. The same model solved more than one hundred other open problems too. These results cover most areas of mathematics broadly. Frontier models are becoming increasingly capable this year. Tens of thousands of mathematical solutions have been generated by AI so far.
Mathematicians Demand Formal Papers Instead of Social Media Posts
Northwestern University mathematician Bryna Kra recalls the August meeting as promising. She notes attendees felt a mixture of excitement and dread about the future. The group asked OpenAI to avoid blog posts or tweets for new findings. They wanted peer-reviewed papers explaining the work fully.
Kra explains that formal papers allow mathematicians to absorb and digest results. This process lets researchers use the findings in their own work. OpenAI ignored this request to publish formally, Kra says. The company prefers social media for quick announcements instead. This approach makes verification much harder for the community.
Traditional scientific processes have been cast aside by these labs. Mathematicians feel the field is a playground for companies like OpenAI and Anthropic. Both firms prepare for blockbuster initial public offerings soon. They rush to outdo each other in solving problems.
Levent Alpöge announced he disproved an 87-year-old conjecture through a tweet. He sent this post shortly after the World Cup final. He did not publish a formal paper for this specific achievement. Kra says math by tweet is not the way to nurture the ecosystem. The data used to train models came from such fertile ground.
Piecemeal announcements are not limited to just the labs in San Francisco or Mountain View. Other researchers also use social media for major claims. This bypasses the standard review and publication steps entirely. It creates confusion about who actually solved what problem.
The Controversy Over Front-Running and Credit Attribution
Tristan Buckmaster, a professor at New York University, accused OpenAI of front-running his work. He had been collaborating with Anthropic employee Levent Alpöge on a personal project. They were using OpenAI tools to help solve an element of the problem. Neither pair had published their work publicly yet.
Buckmaster claimed OpenAI researcher Sébastien Bubeck implied Alpöge should be excluded from any paper. Bubeck suggested this would make things complicated for the group. Buckmaster refused to accept this treatment during negotiations. He threatened to tell the press about his belief that OpenAI had stolen his work.
Meeting notes seen by WIRED show Bubeck worried about what Anthropic is doing. He told a colleague Levent must be talking to leadership right now. The notes suggest Bubeck thought OpenAI could get a Millennium Prize problem easily. There was worry about Anthropic giving all their compute to solve it too.
Bubeck claimed Buckmaster would ruin his career if he spoke out. He said, "If you don't want me to be nice, then I don't have to be nice." This threat appeared in meeting notes according to the report. The situation highlights serious issues with credit attribution in AI research.
Nestor Guillen, a visiting math professor at NYU, tells WIRED there is a perception of mobster behavior from AI companies. He sees this among mathematicians who feel angry about the power accumulation in one place. McCallum disagrees with that characterization of the situation entirely. She insists OpenAI is working responsibly to release results.
Background on the Millennium Prize Problem and AI Agents
The Millennium Prize Problem refers to a legendary million-dollar math challenge. It involves solving complex equations related to fluid dynamics. The Navier-Stokes problem is one specific part of this larger challenge. Solving it would bring a massive prize for the first person or group.
OpenAI deployed thousands of agents to solve this legendary problem in September. They heard rumors that others were closing in on solutions quickly. This triggered a race to find answers before anyone else did. The use of AI agents accelerated the search process significantly.
Tristan Buckmaster had done work toward solving an element of the problem in private. He collaborated with Levent Alpöge outside of formal academic channels. They used OpenAI tools to assist their personal research efforts. This bypassed normal university or lab protocols for collaboration.
The controversy shows how AI agents can outpace human researchers in specific tasks. Companies deploy large numbers of these automated systems without much oversight. It creates a new kind of competition based on compute power rather than ideas alone. Mathematicians worry this undermines the value of their own work.
Community Push for New Tools Like Hexagon and Palomar
Some have set up new tools this year in reaction to the increase in machine-assisted proofs. The idea is to help the community both use new results and sort through what needs further digestion. These tools aim to organize the flood of AI-generated content.
Hexagon is a repository for primarily AI-generated material online. It stores proofs and solutions created by models without human verification yet. Palomar is a registry of machine-verified mathematics instead. It lists work that has passed some form of checking by humans.
These tools help the community use new results more effectively. They allow researchers to filter out low-quality or unverified claims. The goal is to maintain standards in an era of rapid AI output. OpenAI has been directly encouraged to use these tools like Hexagon and Palomar.
Kra notes they have not changed their behavior despite this encouragement from the community. The company continues to release results through blog posts rather than papers. She feels the company's actions are not aligned with the Leiden declaration either. This declaration sets expectations for how AI should interact with mathematical research and learning.
The Leiden declaration calls for AI companies to meet mathematicians' standards fully. More than 4,000 mathematicians signed this call for better collaboration recently. It establishes rules for transparency and proper credit attribution in AI work. OpenAI has not adopted these standards according to many critics.
Why it Matters
Cost, speed, quality, safety, or a new capability drives the urgency of these changes in mathematical research. The current approach threatens the integrity of scientific discovery globally. Mathematicians fear their careers depend on work that companies claim but do not credit properly.
Trust in AI-generated results is eroding among researchers and educators alike. Without formal papers, it becomes hard to verify who actually solved what problem. This lack of verification excludes prior work from other mathematicians from the record. It creates a fragmented history of mathematical progress online.
The accumulation of power in one place causes significant angst within the field. Mathematicians worry about losing control over how their knowledge is presented and used. They want tools that support research rather than replace human judgment entirely.
How OpenSmartRoute helps
A team using OpenSmartRoute gains control when models release unverified results. The router scores every candidate on quality, cost, speed and safety before routing a request. It prevents expensive or unsafe answers by setting hard rules for personal data and regions. A model that answers well gets more traffic while one that fails gets less.
The platform keeps a catalogue with prices and public rankings built from real traffic. An input guard spots prompt injection and personal data before a request leaves the system. 'osr eval' measures routing accuracy on the team's own prompts and can fail a build when it drops. This setup ensures the organization does not rely on unreviewed claims from OpenAI or others.
What to Do
Readers who run models or agents should check if new math solutions are on GitHub soon. Look for official announcements from OpenAI regarding these releases in the coming days. Verify any claims made by social media posts with formal papers when available.
Researchers can use tools like Hexagon and Palomar to track AI-generated proofs. These platforms help sort through what needs further digestion before publication. They provide a way to organize the growing volume of machine-assisted math work.
Mathematicians should demand formal papers for all significant new results from AI labs. This ensures proper credit attribution and allows for peer review of the work. The community must push for standards that protect human researchers from being sidelined by automation.
Supporters of the Leiden declaration can share their calls for better collaboration with tech companies. They can advocate for policies that ensure AI tools enhance rather than replace mathematical research. The future of mathematics depends on how well these boundaries are drawn and respected.
