Scientific & Numerical Languages — Tier 3 Index

Scientific & Numerical Languages — Tier 3 Index

  • Type: Family index (Tier 3)
  • Family: Scientific / numerical computing, statistics, symbolic math, probabilistic programming, array languages, HPC
  • Languages catalogued: 27
  • Last updated: 2026-05-07

Family overview

“Scientific computing” is a tent so big it strains the term. Inside it sit at least five sub-families: commercial numerical environments (MATLAB, Mathematica, Maple, IDL) that were the dominant platforms of the 1990s-2000s; statistics and econometrics (SAS, SPSS, Stata, GAUSS, gretl) where domain users still outnumber R/Python users in places; the APL/array lineage (APL → J, K, Q, A+) that quietly powers high-frequency finance; probabilistic programming (Stan, BUGS/JAGS, Pyro, NumPyro, TFP) where Bayesian methods got a real implementation story; and HPC-oriented languages (Chapel, X10, Modelica, Felix) that tried to displace Fortran+MPI and mostly didn’t. The macro story since ~2015 is Python eating the centre — most new graduate students reach for NumPy/SciPy/PyMC/scikit-learn before MATLAB or SAS — while the frontier-Bayesian, ultra-low-latency-finance, and physical-modelling niches retain their specialised languages.

In our deep library

  • julia — modern open-source numerical/HPC language; the closest thing to a unified successor
  • r — statistics, the dominant academic-stats language since ~2010
  • fortran — still load-bearing in HPC numerical libraries
  • python — the gravitational centre of modern data/scientific work

Tier 3 — the family

LanguageFirst releaseStatus 2026NicheWhy it mattersSource URL
MATLAB1984Active commercial (MathWorks)Engineering, control, signal processingThe lingua franca of EE/control-systems education; Simulink remains unmatched for model-based designhttps://www.mathworks.com/products/matlab.html
GNU Octave1988Active OSSMATLAB-compatibleOpen-source MATLAB-compatible interpreter; lifeline when MATLAB licenses aren’t availablehttps://octave.org/
Scilab1990Active (Dassault)Numerical computingFrench INRIA-origin MATLAB-alike; common in EU engineering schoolshttps://www.scilab.org/
SageMath2005Active OSSComputational mathematicsPython-based front-end binding GAP, PARI, Singular, Maxima, etc. into one notebook environmenthttps://www.sagemath.org/
Wolfram Language / Mathematica1988Active (Wolfram)Symbolic + numeric, knowledgeBest symbolic-math engine in mainstream use; Wolfram Alpha is the same kernel; closed-sourcehttps://www.wolfram.com/mathematica/
Maple1982Active (Maplesoft)Symbolic mathThe other major commercial CAS; popular in engineering math educationhttps://www.maplesoft.com/products/maple/
IDL1977Active (NV5/Harris)Astronomy, atmospheric, remote sensingSolar physics and HST/JWST pipelines historically run on IDL; community now migrating to Pythonhttps://www.nv5geospatialsoftware.com/Products/IDL
Stata1985Active commercialStatistics, econometricsDominant in health economics, biostatistics, and policy research; do-file scripting culturehttps://www.stata.com/
SAS1976Active but decliningEnterprise statisticsPharma clinical-trials gold standard (FDA submissions); losing graduate-student mindshare to R/Pythonhttps://www.sas.com/
SPSS1968Maintenance (IBM)Social-sciences statisticsGUI-driven; entrenched in psychology, sociology, market researchhttps://www.ibm.com/products/spss-statistics
gretl1999Active OSSEconometricsOpen-source econometrics; common in undergraduate teaching as a free Stata alternativehttps://gretl.sourceforge.net/
GAUSS1984Active commercialEconometricsLong-running commercial matrix language for econometrics/finance researchhttps://www.aptech.com/
APL1966NicheArray programmingIverson’s symbolic array language; intellectual ancestor of NumPy, J, K, Qhttps://en.wikipedia.org/wiki/APL_(programming_language)
Dyalog APL1983Active commercialModern APLThe reference modern APL; active community, Dyalog ‘24/’25 conferences; finance + actuarial usershttps://www.dyalog.com/
J1990Active OSSASCII APL successorIverson’s ASCII reformulation of APL; tacit/point-free style; cult followinghttps://www.jsoftware.com/
K1993Active commercial (KX)High-frequency financeWhitney’s terse APL-descendant; the runtime under kdb+; sub-microsecond tick processinghttps://kx.com/
Q2003Active commercial (KX)Readable layer over KEnglish-keyword surface for K; the language analysts actually write at trading firmshttps://code.kx.com/q/
A+1988HistoricalMorgan Stanley APL variantIn-house Morgan Stanley fork of APL; mostly displaced internally; open-sourced as historical artifacthttp://www.aplusdev.org/
kona2005Maintenance OSSOpen-source K3OSS reimplementation of K3; pedagogical / hobbyisthttps://github.com/kevinlawler/kona
Stan2012ActiveProbabilistic programming, HMCReference implementation of Hamiltonian Monte Carlo / NUTS; Columbia University; the serious Bayesian’s compilerhttps://mc-stan.org/
BUGS / OpenBUGS / WinBUGS1989Maintenance / historicalBayesian Gibbs samplingThe original Bayesian DSL; spawned an entire generation of applied Bayesian work; superseded by JAGS/Stanhttps://www.mrc-bsu.cam.ac.uk/software/bugs/
JAGS2007Active OSSBayesian Gibbs sampling”Just Another Gibbs Sampler”; cross-platform BUGS-compatible engine; popular with ecologistshttps://mcmc-jags.sourceforge.io/
Pyro / NumPyro2017 / 2019ActivePython-embedded probabilisticUber’s PyTorch-based Pyro; NumPyro is the JAX-based variant from the same team — fast HMC for modern ML researchershttps://pyro.ai/
TensorFlow Probability2018ActiveTF-embedded probabilisticGoogle’s TF/JAX-based probabilistic stack; tight Keras/JAX integrationhttps://www.tensorflow.org/probability
Modelica1997Active (Modelica Assoc.)Physical-system modelingAcausal, equation-based modeling for multiphysics (electrical+mechanical+thermal+control); used in Dymola, OpenModelica, Wolfram SystemModelerhttps://modelica.org/
Felix2002Niche/researchScientific systems languageHigh-performance functional/scientific language; small but persistent research communityhttps://felix-lang.org/
Chapel2009Active (HPE/Cray)HPC PGASPartitioned global address space; data-parallel + task-parallel; Cray’s bet at a productive HPC successor to Fortran+MPIhttps://chapel-lang.org/
X102004Maintenance/researchHPC PGASIBM’s PGAS language for the DARPA HPCS program; APGAS runtime survived as a research vehiclehttp://x10-lang.org/

Notable threads

Python ate the centre but the periphery survives. The single biggest story in scientific computing 2010-2026 is the migration of general-purpose users to Python: NumPy/SciPy/Pandas/Matplotlib/Jupyter became the default toolkit for new students in physics, biology, chemistry, ML, and most of engineering. MATLAB held its ground in EE/controls (because Simulink has no real open-source equivalent), Stata held its ground in health economics (because reviewer culture demands it), SAS held its ground in pharma clinical trials (because FDA submission templates are SAS-flavoured), and Wolfram/Maple held their ground in symbolic math (because nobody has matched their CAS engines). Everything else has shifted Python-ward, with julia picking up niche converts who want speed without leaving the high-level abstraction tier.

The K/Q/kdb+ finance niche. A handful of languages — K, Q, J, A+, APL — remain quietly load-bearing at trading firms. The pattern is consistent: terse array operations on columnar in-memory data, sub-microsecond latencies, and a small priesthood of practitioners willing to read code that looks like line noise. KX’s kdb+ runs essentially every major exchange’s tick database. Most software developers will never see these languages; most quant engineers at top firms will write Q daily. They survive because the alternatives (NumPy/Pandas, Polars, even DuckDB) still can’t match raw kdb+ throughput on the workloads it was designed for.

Probabilistic programming finally worked. From the 1990s through the early 2010s, “Bayesian methods” mostly meant WinBUGS notebooks and a lot of patience. The 2012 Stan release of HMC/NUTS made full-Bayesian inference tractable on modern problems for the first time. Pyro (2017) and NumPyro (2019) brought the same machinery into PyTorch and JAX, letting ML researchers actually compose neural networks with probabilistic models. As of 2026 a graduate student doing applied Bayesian work picks Stan (for serious published inference), brms/rstan (for the R ecosystem), or NumPyro (for differentiable models). BUGS/JAGS persist mostly as teaching tools and in ecology where the historical literature uses them.

The HPC-language graveyard. Chapel, X10, Fortress, Cilk Plus, Sisal, ZPL, HPF — the 2000s-2010s saw a parade of attempts at “the language that replaces Fortran+MPI for exascale.” None succeeded. The actual exascale codes shipping on Frontier, Aurora, and El Capitan are written in C++/Fortran with Kokkos/RAJA/OpenMP-target/SYCL (cross-link gpu-and-shaders). Chapel survives at HPE as a research/productivity vehicle; X10 became a research artifact. The lesson the field absorbed: replacing Fortran is much harder than building libraries on top of it.

Modelica and the equation-based niche. Modelica is unusual in this list because it’s not really a programming language — it’s an equation language. You write the physics (KCL, mechanical balance, thermal flow) and the compiler does index reduction, symbolic manipulation, and DAE solving. It powers Dymola (Dassault), Wolfram SystemModeler, and OpenModelica, and has carved out a durable niche in automotive, aerospace, and HVAC system simulation where Simulink’s signal-flow paradigm is the wrong abstraction.

Citations