LinearSolveBench
www.autodidakt.ai- Category
- Developer Tools
- Rank
- No. 2758Tools index
Previous survey · No. 2111 ·
- Type
- TOOL
- Use case
- Model & Agent Evaluation
- GitHub
- 1 stars
- Date
About
LinearSolveBench is a benchmark that measures how well an AI model or coding harness can write fast, accurate, and general numerical solvers for large sparse linear systems of equations in C. Its stated goal is to encourage algorithmic progress in numerical linear solver design.
At a glance
LinearSolverBench evaluates whether an AI system can produce fast, accurate C solvers for large sparse systems. It is worth using if your question is about numerical solver discovery and you can compare runs under consistent conditions.
Good fit
Benchmark authors and numerical computing researchers evaluating agent-generated solvers.
Consider something else when
Choose an existing numerical library if you need a production solver for one application.
Start here
Read the scoring rules and choose one matrix family. Compare a candidate with the reference under the same build and runtime conditions, checking accuracy before interpreting speed.
Cost and requirements
Apache 2.0 for the project, with separate attribution for some matrices. The documented setup needs Python 3.12, a C or C++ compiler, CMake, Ninja, Git and nm.
Limits to know
Results depend on the matrix family, hardware and scoring rules. Isolate generated native code and report the tested cases when comparing agents.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Gives researchers a way to measure whether coding agents can produce genuinely fast, correct numerical solvers rather than just passing generic coding tests.
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.