Jimmy Xin

George Tsoukalas, Jasper Lee, John Jennings et al.

We present PutnamBench, a new multi-language benchmark for evaluating the ability of neural theorem-provers to solve competition mathematics problems. PutnamBench consists of 1692 hand-constructed formalizations of 640 theorems sourced from the William Lowell Putnam Mathematical Competition, the premier undergraduate-level mathematics competition in North America. All the problems have formalizations in Lean 4 and Isabelle; a substantial subset also has Coq formalizations. PutnamBench requires significant problem-solving ability and proficiency in a broad range of topics taught in undergraduate mathematics courses. We use PutnamBench to evaluate several established neural and symbolic theorem-provers. These approaches can only solve a handful of the PutnamBench problems, establishing the benchmark as a difficult open challenge for research on neural theorem-proving. PutnamBench is available at https://github.com/trishullab/PutnamBench.

Janine Lohse, Tim Rohde, Jimmy Xin et al.

There has recently been exciting progress in the realm of *probabilistic separation logics*. An important subclass of these -- including PSL, Lilac, Bluebell, and pcOL -- are *general-purpose probabilistic logics* (or GPLs, for short), meaning that they provide primitive Hoare-style assertions about probability distributions on the program state, along with powerful modularity principles like *independence* and *conditioning*. However, none of these logics support reasoning about dynamically allocated memory (i.e., pointers into a heap), let alone the more sophisticated resource algebra-based ghost state of modern separation logics like Iris. We argue that this is due to a fundamental obstacle: since the shape of memory (and identity of memory locations) may differ under different random outcomes, it is unclear how pointer ownership can be harmonized with probabilistic independence and conditioning. Furthermore, none of the existing GPLs have been mechanized in a proof assistant. In this paper, we take a substantial first step towards a marriage of GPLs and modern separation logics like Iris, in the form of **Amaryllis**. Amaryllis is the first GPL to support independence and conditional reasoning while also handling dynamic memory allocation. To overcome the aforementioned obstacle, we propose a new *indexed valuation*-style model of probabilistic assertions, whereby ownership of a standard Iris-style resource (e.g., heaps) can be promoted to ownership at the level of distributions in a generic fashion. We then show how to adapt the central Iris notions of *frame-preserving update*, *authoritative resource algebras*, and the *weakest precondition modality* to be sound for probabilistic reasoning and validate dynamic allocation. Finally, we have mechanized all our results in the Rocq proof assistant and developed an Iris-based proof mode for conducting proofs within Amaryllis.

Jimmy Xin, Alex Schneidman, Chris Cummins et al.

We present AXLE (Axiom Lean Engine), a cloud service for Lean 4 proof manipulation, extraction, and verification. Recent progress in AI for mathematics -- reinforcement learning pipelines, agentic proving workflows, dataset curation -- demands Lean 4 tooling that scales to millions of requests while remaining correct and robust; existing infrastructure offers parallel compilation but not scalable proof verification, higher-level proof manipulation, multi-version support, or per-request isolation at the throughput modern AI workflows require. AXLE provides 14 Lean 4 metaprogramming tools spanning strict proof verification, declaration metadata extraction, semantic source manipulation, deterministic proof repair and simplification, and lemma extraction. The service runs as a multi-tenant cloud deployment with per-request isolation and concurrent support for multiple Lean 4 and Mathlib versions, accessible via a Python SDK, command-line interface, web UI, MCP server, and raw HTTP API. AXLE is publicly available and free to use at https://axle.axiommath.ai and via the axiom-axle PyPI package, with no local Lean 4 installation required. It has served over 500 million requests to date and is the underlying infrastructure for Axiom Math's proving efforts, including its 12/12 score on the 2025 Putnam competition.