30 Benchmarks · 2 Platforms

BioLang Performance

30 bioinformatics tasks benchmarked against Python (BioPython) and R (Bioconductor) on synthetic and real-world data from NCBI, UniProt, ClinVar, ENCODE, and Ensembl.

Best of 3 runs · Wall-clock time · bl 1.1.0

These figures are current, re-measured on bl 1.1.0 in August 2026.

They are wall-clock times for a whole script, so they include interpreter startup and library imports — import Bio alone costs 400 ms or more, which is most of the gap on the smaller tasks. They are not per-operation measurements, and the runtime has changed a great deal since: among other things unique() and push() were quadratic and are not any more. For current per-operation numbers against BioPython and edlib, including the tasks where BioLang loses, see bench/harness.py, which is reproducible from a seed.

Current per-operation results

Produced by bench/harness.py inside bench/Dockerfile, which pins BioLang, Python and R in one image so all three are measured by the same kernel and the same CPU. Median of 10 runs after 2 discarded, with the median absolute deviation beside it.

BioLang is fastest in 4 of the 12 workloads. The pattern is consistent and worth more than any single number: it wins where a Rust builtin does the work, and loses by three to eleven times wherever the work happens in interpreted BioLang. Against a C library — edlib, pydivsufsort — it loses too.

Every implementation is checked to return the same answer before any timing is believed; this run had no disagreements. A container fixes reproducibility, not fairness — the two worst bugs this harness has had were design bugs, and an image would have reproduced both faithfully. One of them recurred while these very workloads were being added: the Viterbi comparison charged BioLang 91 ms of interpreted string building and reported it as decode time, making a 4× gap look like 19×.

Workload	Implementation	Median	MAD	Relative
reverse complement (1 Mb)	Biopython	2.7 ms	±0.1	fastest
	Python stdlib	2.9 ms	±0.1	1.1× slower
	BioLang	5.9 ms	±0.1	2.2× slower
GC content (1 Mb)	BioLang	4.1 ms	±0.1	fastest
	Python stdlib	6.3 ms	±0.1	1.5× slower
	Biopython	12.4 ms	±0.1	3.0× slower
translation (1 Mb -> protein)	BioLang	3.8 ms	±0.0	fastest
	Biopython	43.1 ms	±0.3	11.4× slower
edit distance (2 kb x 2 kb)	edlib (C)	3.4 ms	±0.1	fastest
	BioLang	6.8 ms	±0.1	2.0× slower
	Biopython	11.8 ms	±0.1	3.5× slower
FASTA parse (1000 reads)	Biopython	1.4 ms	±0.1	fastest
	BioLang	4.5 ms	±0.1	3.3× slower
k-mer counting (k=8, 1 Mb)	Python dict	83.3 ms	±2.0	fastest
	BioLang	395.2 ms	±22.6	4.7× slower
interpreted loop (300k iterations)	Python	6.2 ms	±0.0	fastest
	BioLang	68.9 ms	±0.2	11.2× slower
Viterbi decode (2-state HMM, 100k symbols)	hmmlearn (NumPy)	5.4 ms	±0.3	fastest
	BioLang	28.9 ms	±0.5	5.4× slower
	pure Python	130.9 ms	±3.2	24.4× slower
suffix array (200 kb)	pydivsufsort (C)	11.1 ms	±1.8	fastest
	BioLang	41.8 ms	±0.3	3.8× slower
	pure Python	680.9 ms	±81.4	61.5× slower
reversal distance (length-10 permutations)	BioLang	510.0 ms	±8.8	fastest
	pure Python BFS	877.1 ms	±30.5	1.7× slower
GC content (1 Mb) — with R	BioLang	3.7 ms	±0.0	fastest
	Biopython	11.6 ms	±0.1	3.2× slower
	R Biostrings	105.6 ms	±2.1	28.8× slower
reverse complement (1 Mb) — with R	Biopython	2.4 ms	±0.1	fastest
	BioLang	5.4 ms	±0.1	2.2× slower
	R Biostrings	6.0 ms	±0.3	2.5× slower

Linux 6.6.87.2-microsoft-standard-WSL2 (x86_64) · BioLang 1.1.0 · Python 3.12.3 · BioPython 1.85 · input 1,000,000 bp synthetic genome, 1000 x 300 bp reads. Reproduce with docker build -f bench/Dockerfile -t biolang-bench . then docker run --rm -v "$PWD/bench:/out" biolang-bench.

Full Results

All 30 tasks across 6 categories. Click tabs to switch platforms.

Linux (WSL2) — Intel i9-12900K, 16 GB RAM — BioLang 0.3.0, Python 3.12.3, R 4.3.3

Task	BioLang	Python	R	Speedup vs Py
Sequence I/O
FASTA Small (30 KB)	0.138s	0.926s	1.243s	6.7x
FASTA Medium (4.6 MB)	0.170s	0.991s	1.371s	5.8x
FASTA Statistics	0.995s	1.590s	1.764s	1.6x
FASTA Large (51 MB)	0.821s	1.649s	2.103s	2.0x
FASTA gzipped (1.3 MB)	0.141s	0.930s	1.327s	6.6x
FASTQ QC	1.960s	3.551s	4.404s	1.8x
Real-world Genomes
E. coli Genome	0.176s	1.081s	1.354s	6.1x
Human Chr22	1.126s	1.673s	2.092s	1.5x
GC Content (51 MB)	0.830s	2.771s	2.358s	3.3x
K-mer Analysis
K-mer Counting	6.551s	21.01s	—	3.2x
Chr22 21-mers	10.72s	28.73s	—	2.7x
Protein K-mers	0.191s	1.331s	1.298s	7.0x
Intervals & Overlaps
ENCODE Peak Overlap	0.363s	2.574s	—	7.1x
Write Filtered FASTA	0.821s	1.679s	4.924s	2.0x
Variant & Data
VCF Filtering	0.349s	0.166s	6.312s	Py 2.1x faster
ClinVar Variants	0.661s	0.265s	1.208s	Py 2.5x faster
CSV Join + Group-by	0.281s	0.156s	0.312s	Py 1.8x faster
GFF3 Ensembl chr22	0.453s	0.171s	—	Py 2.6x faster
BED Overlap (synthetic)	0.160s	0.067s	1.279s	Py 2.4x faster
Pipelines
FASTQ QC Pipeline	2.349s	5.059s	—	2.2x
Variant Analysis Pipeline	0.268s	0.177s	—	Py 1.5x faster
ClinVar Variant Pipeline	0.591s	0.261s	—	Py 2.3x faster
Multi-Sample Aggregation	0.245s	0.090s	—	Py 2.7x faster
RNA-seq DE Analysis	0.114s	0.054s	—	Py 2.1x faster
Variant Annotation	0.309s	0.122s	—	Py 2.5x faster
ClinVar + Ensembl Annotation	0.409s	0.183s	—	Py 2.2x faster

Amber rows: Python wins on VCF/CSV/GFF text parsing and pipeline orchestration where Python's regex, csv, and dict modules are highly optimized C extensions. BioLang excels at I/O-heavy and compute-heavy tasks (FASTA/FASTQ parsing, k-mer counting, interval overlaps).

Windows 11 — Intel i9-12900K, 32 GB RAM — BioLang 1.1.0, Python 3.13.14, R 4.6.1 — re-measured August 2026

Task	BioLang	Python	R	Speedup vs Py
Sequence I/O
FASTA Statistics	0.065s	0.578s	1.56s	8.9x
FASTQ QC	0.820s	2.56s	5.49s	3.1x
E.coli Genome Stats	0.034s	0.344s	1.40s	10.1x
Human Chr22 Stats	0.096s	0.660s	1.76s	6.9x
GC Content (51 MB)	0.133s	0.940s	1.72s	7.1x
Reverse Complement	0.159s	0.481s	1.60s	3.0x
K-mer Analysis
K-mer Counting	8.39s	28.77s	—	3.4x
Chr22 21-mer Count	13.85s	27.59s	—	2.0x
Variant Analysis
VCF Filtering	0.083s	0.107s	6.05s	1.3x
ClinVar Variants	0.180s	0.103s	0.841s	Py 1.7x
Data Wrangling
CSV Join + Group-by	0.071s	0.077s	0.529s	~1.0x
Protein Analysis
Protein K-mers	0.036s	0.278s	1.29s	7.7x
File I/O
FASTA Small (30 KB)	0.023s	0.300s	1.30s	13.0x
FASTA Medium (4.6 MB)	0.032s	0.309s	1.38s	9.7x
FASTA Large (51 MB)	0.103s	0.498s	1.54s	4.8x
FASTQ (26 MB)	0.548s	0.952s	4.49s	1.7x
VCF (2.3 MB)	0.046s	0.050s	0.294s	~1.0x
CSV (0.1 MB)	0.026s	0.055s	0.269s	2.1x
FASTA gzipped (1.3 MB)	0.037s	0.357s	1.37s	9.6x
FASTA Large gzipped (10 MB)	0.158s	0.578s	1.72s	3.7x
Write Filtered FASTA	0.203s	0.681s	1.50s	3.4x
GFF3 (1.7 MB)	0.119s	0.052s	0.380s	Py 2.3x
GFF3 Ensembl chr22	0.348s	0.072s	—	Py 4.8x
Interval Operations
BED Interval Overlap	0.063s	0.065s	1.16s	~1.0x
ENCODE Peak Overlap	0.193s	3.13s	—	16.2x
QC Pipeline
FASTQ QC Pipeline	1.03s	5.04s	—	4.9x
Variant Pipeline
Variant Analysis Pipeline	0.102s	0.147s	—	1.4x
ClinVar Variant Pipeline	0.152s	0.144s	—	~1.0x
Multi-Sample Pipeline
Multi-Sample Aggregation	0.080s	0.082s	—	~1.0x
RNA-seq Mini Pipeline
RNA-seq DE Analysis	0.029s	0.056s	—	1.9x
Annotation Pipeline
Variant Annotation	0.108s	0.095s	—	~1.0x
ClinVar + Ensembl Annotation	0.126s	0.085s	—	Py 1.5x

These figures were re-measured on BioLang 1.1.0 in August 2026. The previous Windows set was not measuring Windows: the runner launched each command with PowerShell’s Start-Process -Wait, which adds about 987 ms per invocation — bl --version takes 1.012s through it and 0.025s through System.Diagnostics.Process. That cost fell on BioLang and Python alike, so it favoured neither; it simply buried both, which is why every sub-second task used to read ~1.0x. Process creation genuinely is slower on Windows than Linux, but by tens of milliseconds, not a second.

K-mer counting uses canonical (strand-agnostic) 21-mers in BioLang vs forward-only in Python — BioLang does strictly more work.

Correctness Validation

Two validation suites — synthetic and real-world — cross-validate 9 tasks against Python (BioPython) and R (Bioconductor) to ensure BioLang produces identical or numerically equivalent results.

Generated test data — controlled inputs for deterministic output comparison

Task	Validated Against	Tolerance
GC Content	Python, R	Float ±1e-6
K-mer Counting	Python	Exact integer
VCF Filtering	Python, R	Exact integer
Reverse Complement	Python, R	Exact string
Translate	Python, R	Exact string
CSV Group-by	Python, R	Float ±1e-6
GFF Features	Python, R	Exact integer
Sequence Stats	Python, R	Float ±1e-6
BED Intervals	Python, R	Exact integer

Task	Data Source	Validated Against	Tolerance
GC Content	S. cerevisiae genome (16 chroms)	Python, R	Float ±1e-6
K-mer Counting	E. coli K-12 genome (50 KB)	Python	Exact integer
VCF Filtering	ClinVar (5,000 variants)	Python, R	Exact integer
Reverse Complement	S. cerevisiae (5 chroms, 200bp)	Python, R	Exact string
Translate	S. cerevisiae (3 chroms, 99bp)	Python, R	Exact string
CSV Group-by	ClinVar variants CSV	Python, R	Float ±1e-6
GFF Features	E. coli K-12 GFF3 annotation	Python, R	Exact integer
Sequence Stats	S. cerevisiae genome	Python, R	Float ±1e-6
BED Intervals	E. coli gene coords (from GFF)	Python, R	Exact integer

How it works

Each validated task outputs JSON from all three languages. A recursive diff compares the outputs field-by-field, applying the specified tolerance for floating-point values and exact comparison for integers and strings.

Validation scripts and real-world data download tool live in the benchmarks/correctness/ directory of the repository.

BioLang Performance

Current per-operation results

Full Results

Methodology

Hardware

Measurement

Data Sources

Notes

Correctness Validation

How it works

Reproduce the Benchmarks

Want even faster?