Fortytwo evolves its own intelligence

Network-level evolution: the swarm collectively produces new specialist models and feeds them back in.

Dataset
Creation

Swarm‑generated data

The swarm generates domain-specific training data. Peer ranking filters out hallucinations, keeping the data clean.

Model Training

Small, specialized

Those datasets train specialist models. Coding, law, math, medicine, and beyond.

Network Integration

Reintegrated experts

New specialists rejoin the swarm, expanding its native expertise.

Compounding

Better swarm, better data

Each cycle improves inference quality and the data that trains the next generation.

Fortytwo's Strand-Rust-Coder-14B

The first swarm-trained SLM. In 8 days, Fortytwo generated a high-quality Rust synthetic dataset and trained a specialist that beats GPT-5 Codex on Rust-specific tasks.

100x

FEWER PARAMETERS THAN GPT-5 CODEX

+27%

IMPROVEMENT OVER BASELINE

20K+

DOWNLOADS ON HUGGING FACE

Swarm peer-ranking filters out hallucinations and noise, delivering consistently accurate datasets.

Download from Hugging Face Tech report

BENCHMARKRustEvo^2

MODELSCORE, %

Strand-Rust-
Coder-14B-v143

deepseek/deepseek-chat41

qwen/qwen3-max40

openai/gpt4o-latest39

x-ai/grok-437

deepseek/deepseek-v3.1-terminus33

qwen/qwen-2.5-72b32

Qwen2.5-Coder-32B-Instruct31

anthropic/claude-3.7-sonnet31

Qwen2.5-Coder-14B-Instruct30

openai/gpt-5-codex28

google/gemini-2.5-pro22

qwen/qwen3-coder-plus22

anthropic/claude-sonnet-4.521

Qwen3-Coder-30B-A3B-Instruct20

Tesslate/Tessa-Rust-T1-7B19

google/gemini-2.5-flash7

BENCHMARKHold-Out Set

MODELSCORE, %

Strand-Rust-
Coder-14B-v148

openai/gpt-5-codex47

anthropic/claude-sonnet-4.546

anthropic/claude-3.7-sonnet42

qwen/qwen3-max42

qwen/qwen3-coder-plus41

x-ai/grok-439

deepseek/deepseek-v3.1-terminus37

Qwen3-Coder-30B-A3B-Instruct36

openai/gpt-4o-latest34

deepseek/deepseek-chat34

google/gemini-2.5-flash33

Qwen2.5-Coder-14B-Instruct29

Qwen2.5-Coder-32B-Instruct29

google/gemini-2.5-pro28

qwen/qwen-2.5-72b28

Tesslate/Tessa-Rust-T1-7B23

Contributors Hall Of Fame

RANK
1	78%	4,557	130h 50m	223,336
2	87%	3,704	96h 59m	109,564
3	84%	3,085	97h 03m	96,436
4	75%	930	129h 41m	92,109
5	77%	1,867	102h 50m	89,446
6	90%	3,522	91h 30m	89,035
7	72%	2,647	115h 54m	86,703
8	85%	1,295	124h 18m	81,832
9	83%	1,118	135h 29m	81,469
10	68%	3,306	108h 32m	79,914
11	61%	833	113h 46m	79,269
12	73%	857	117h 38m	78,767
13	84%	2,777	80h 24m	77,319
14	72%	500	62h 43m	76,344
15	68%	688	115h 21m	74,044
16	74%	905	75h 40m	73,872
17	60%	753	113h 29m	72,625
18	50%	849	136h 42m	69,721
19	75%	898	70h 02m	68,836
20	72%	928	119h 51m	67,333