By verification grade
8 of 32 carry an independent check. Open the review queue.
Open longest before falling
- 57 yrMatrix multiplication exponent below 2.3710449, machine-checked over every commutative ringposed 1969 · GPT-6 Astra, Claude Fable 5.1, Claude Opus 5 and Opus 5.5; Meta Muse Spark 1.3 for the final numerical search
- 57 yrMatrix multiplication exponent bounded by 9/4posed 1969 · Unreleased internal OpenAI model, not named in the release
- 57 yrMatrix multiplication exponent lowered to below 2.371177posed 1969 · AlphaEvolve
- 56 yr4×4 complex matrix multiplication in 48 scalar multiplicationsposed 1969 · AlphaEvolve (Gemini-based)
From the 21 of 32 entries recording a posed year. Add a missing one.
| Date | Finding | Lab | Model | Verification | Autonomy | Field |
|---|---|---|---|---|---|---|
| 2026-10-27 | Weak normalization implies strong normalization in pure type systems with βη-conversion, proved in Lean | Independent | Claude Opus 5.5 in Claude Code, adapting an OpenAI Lean proof of the β case | Formally verified | Collaborative | computer-science |
| 2026-10-26 | A subexponential algorithm for counting 3-SAT solutions, refuting the Exponential Time Hypothesis | Anthropic | Unreleased internal Anthropic research model, not named or versioned; the paper's author line is Claude | Formally verified | AI-led | computer-science |
| 2026-10-26 | A subexponential counting algorithm for fixed-width CNF, refuting the Exponential Time Hypothesis | OpenAI | Unreleased internal OpenAI model, not named in the release | Claimed | AI-led | computer-science |
| 2026-10-21 | PHP2Shell: two critical PHP remote code execution flaws found with AI assistance | Calif / Wiz Research / OpenAI | Gemini 4 Argon and GPT-5.6 Sol, with the Wiz Atlas tool; which model did what differs by source, see caveats | Claimed | AI-assisted | computer-science |
| 2026-10-16 | NIST elliptic-curve seeds traced to a 1990s C random number generator seeded with the clock | Anthropic | Internal Anthropic research model (Claude), not named in the announcement | Claimed | AI-led | computer-science |
| 2026-10-15 | LLM-designed quantum circuits with fewer gates and qubits than published designs in seven of eight tasks | Anthropic | Claude (the announcement names Claude Mythos 5) | Claimed | Search scaffold | computer-science |
| 2026-10-12 | Lower bound of 21 multiplications for 3×3 matrix multiplication over every field, with a Lean proof | Independent | OpenAI models through ChatGPT and Codex, versions not recorded in the repository (GPT-6 Astra and GPT-6 Sol by the author's later account to vibemathed) | Claimed | Collaborative | computer-science |
| 2026-10-07 | Matrix multiplication exponent below 2.3710449, machine-checked over every commutative ring | Independent | GPT-6 Astra, Claude Fable 5.1, Claude Opus 5 and Opus 5.5; Meta Muse Spark 1.3 for the final numerical search | Formally verified | AI-led | computer-science |
| 2026-10-07 | Counterexamples to Zak's modified integer round-down conjecture for the skiving stock problem | Independent | Claude Fable 5.1 and GPT-5.6 Sol Pro | Claimed | Collaborative | computer-science |
| 2026-10-06 | Matrix multiplication exponent bounded by 9/4 | OpenAI | Unreleased internal OpenAI model, not named in the release | Formally verified | AI-led | computer-science |
| 2026-10-06 | Integer multiplication below n log n, against the Schönhage–Strassen optimality conjecture | OpenAI | Unreleased internal OpenAI model, not named in the release | Claimed | AI-led | computer-science |
| 2026-10-05 | Truly subquadratic 3SUM and truly subcubic APSP, refuting two central fine-grained complexity hypotheses | Anthropic | Unreleased internal Anthropic research model (Claude), not named in the paper | AI-led | computer-science | |
| 2026-09-30 | Six-task counterexample to the Kernel Conjecture of pinwheel scheduling | Independent | GPT-6 Astra (in OpenAI Codex) | Formally verified | AI-led | computer-science |
| 2026-09-15 | Exponential lower bound for the bit pigeonhole principle in unrestricted resolution over parities | Independent | GPT-6 Astra (mathematics, version 1 formalization and drafting); Claude Fable 5.1 and Claude Opus 5 (revision 1 formalization) | Claimed | AI-led | computer-science |
| 2026-09-02 | A quantum oracle separating QMA(2) from QMA | MIT / Columbia / University of Washington | ChatGPT 5.6 Sol | AI-assisted | computer-science | |
| 2026-08-31 | Unrestricted Boolean multiplicative complexity of four-term binary polynomial multiplication | Independent | GPT-5.6 Sol; Claude Opus 5 | Formally verified | AI-assisted | computer-science |
| 2026-08-29 | Randomized metric distortion improved to 2.3282 | Independent | GPT-5.6 Sol; Claude Opus 5.0 | AI-led | computer-science | |
| 2026-08-26 | Rapid mixing for spin systems on graphs of girth at least five | Independent | GPT-5.6 Sol Ultra | Collaborative | computer-science | |
| 2026-08-24 | Optimal linear encoding rate for lossy compression of Bernoulli sources | Independent | GPT-5.6 Sol | AI-led | computer-science | |
| 2026-08-20 | Marton's inner bound shown not to reach the broadcast channel capacity region | Independent | GPT-5.6 Sol, Claude Fable 5 and Claude Opus 5 | AI-assisted | computer-science | |
| 2026-08-19 | Counting linear extensions below the 2ⁿ barrier | Independent | Claude Opus 5; ChatGPT 5.6 Sol | AI-led | computer-science | |
| 2026-08-17 | Matrix multiplication exponent lowered to below 2.371177 | Google DeepMind | AlphaEvolve | Search scaffold | computer-science | |
| 2026-08-11 | A prescribed Hamiltonian cycle that a book-embedding algorithm cannot produce | Independent | OpenAI Codex, with Claude for adversarial review | Formally verified | AI-assisted | computer-science |
| 2026-07-28 | Kemeny rank aggregation shown NP-hard for three voters | Independent | GPT-5.6 Sol Ultra, Claude Fable 5 | Formally verified | AI-led | computer-science |
| 2026-07-22 | Counterexample to the Dinitz–Garg–Goemans conjecture | Independent | GPT-5.6 Pro | Formally verified | AI-led | computer-science |
| 2025-11-20 | Early science acceleration experiments with GPT-5 | OpenAI | GPT-5 | Peer reviewed | AI-assisted | computer-science |
| 2025-09-25 | Limits to black-box amplification in QMA, with the key step written by GPT-5 | UT Austin / CWI Amsterdam / University of Bonn | GPT-5 Thinking | Peer reviewed | AI-assisted | computer-science |
| 2025-05-14 | 4×4 complex matrix multiplication in 48 scalar multiplications | Google DeepMind | AlphaEvolve (Gemini-based) | Independently checked | Search scaffold | computer-science |
| 2025-03-12 | A fully machine-generated paper passed workshop peer review | Sakana AI | The AI Scientist-v2 | AI-led | computer-science | |
| 2023-12-14 | Improved heuristics for online bin packing | Google DeepMind | FunSearch (PaLM 2 / Codey) | Peer reviewed | Search scaffold | computer-science |
| 2023-06-07 | Faster sorting routines discovered and merged into the LLVM C++ library | Google DeepMind | AlphaDev (AlphaZero-based) | Independently checked | Search scaffold | computer-science |
| 2022-10-05 | Faster matrix-multiplication algorithms found by reinforcement learning | Google DeepMind | AlphaTensor (AlphaZero-based) | Peer reviewed | Search scaffold | computer-science |