Boole-1.7B is a 1-bit language model derived from Qwen3-1.7B with 247 MB of stored weights. It achieves 85.9% of the baseline's performance with a 13.9× reduction in weight storage size, scoring 50.55 across ten benchmarks.

What makes 1-bit quantization difficult?

Two main challenges make 1-bit quantization hard: sharing one magnitude scale across many weights discards information about different weight magnitudes, and the scale values themselves require additional storage, adding overhead to the model size.

How does Boole-1.7B compare to other 1-bit models?

At nearly the same weight storage as binary Bonsai-1.7B, Boole averages 5.58 points higher (50.55 vs 44.97). Ternary Bonsai-1.7B scores 1.02 points higher but requires 1.9× more storage. Boole has the highest intelligence density among all evaluated models at 2.85 per GB.

What benchmark tasks were used to evaluate Boole-1.7B?

Evaluation uses an equal-weight mean of ten tasks: GSM8K, MATH-500, HumanEval+ and MBPP+, MMLU-Redux, GPQA-Diamond, MuSR, IFEval, IFBench, and BFCLv3. Answers are extracted by rule-based parsers with language-model fallback when parsing fails.

In which tasks does Boole-1.7B show the largest improvements over Bonsai?

Boole improves on binary Bonsai in seven of ten tasks, with the largest gains in math and code, while its scores are lower on MuSR, IFEval, and function calling (BFCLv3).

Open source AI

Boole-1.7B

A 1-bit language model with 247 MB of stored weights

Try precisionnumerics.ai (new tab)0 votes. Log in to give Boole-1.7B a nod

Community votes: 0
DR by Ahrefs: 2 −1 (down) since last check

Ahrefs Domain Rating benchmarked 27 Oct 2026, rebuilt 26 Oct 2026.

A 1-bit quantized language model derived from Qwen3-1.7B that achieves 85.9% of baseline performance with 13.9× smaller weight storage.

What it is

Screenshot of the Boole-1.7B homepage — precisionnumerics.ai, as visitors see it

Boole-1.7B is a 1-bit language model derived from Qwen3-1.7B, released by Precision Numerics. It uses binary weights (+1 or −1) with shared magnitude scales to achieve extreme compression, storing 247 MB of weights compared to 3441 MB for the BF16 baseline. The model scores 50.55 across ten benchmarks, retaining 85.9% of the baseline's performance while providing a 13.9× reduction in stored weight size.

The model is intended for English-language inference in non-thinking mode. Among small language models below 3 GB of weight storage, Boole-1.7B has the highest benchmark average at its storage size or below, with an intelligence density of 2.85 per GB—the highest of all evaluated models. It improves on binary Bonsai-1.7B in seven of ten tasks, with the largest gains in math and code.

What it is built with

247 MB stored weights
50.55 benchmark average across ten tasks
85.9% of baseline performance
13.9× reduction in stored weight size
Highest intelligence density at 2.85 per GB
Binary weights with shared magnitude scales
Improvement on binary Bonsai in seven of ten tasks
English-language inference in non-thinking mode
Available in multiple formats: Safetensors and GGUF
Pareto-optimal efficiency frontier

Pricing

Pricing model: Open source

Boole-1.7B metrics

Benchmarks for precisionnumerics.ai: Ahrefs Domain Rating and Majestic link data, with the change since the last build.
Metric	Value
Domain Rating	2 −1 (down) since last check
Trust Flow	0 no change since last check
Citation Flow	21 no change since last check
Referring domains	26 no change since last check
External backlinks	69 no change since last check

Ahrefs Domain Rating benchmarked 27 Oct 2026, rebuilt 26 Oct 2026. Majestic benchmarked 25 Oct 2026, rebuilt 25 Oct 2026. Link data built by Majestic. Trust Flow, Citation Flow and Topical Trust Flow are trademarks of Majestic-12 Ltd.

Boole-1.7B domain build sheet

Open the build sheet

Public RDAP record, last read 26 Oct 2026, 22:12 UTC.
Domain age	2 months
Registered	23 Aug 2026
Expires	23 Aug 2028
Registrar	Porkbun LLC
DNSSEC	Not signed

Inspect precisionnumerics.ai's registration

Compare and explore

Questions from the bench

What is Boole-1.7B?: Boole-1.7B is a 1-bit language model derived from Qwen3-1.7B with 247 MB of stored weights. It achieves 85.9% of the baseline's performance with a 13.9× reduction in weight storage size, scoring 50.55 across ten benchmarks.
What makes 1-bit quantization difficult?: Two main challenges make 1-bit quantization hard: sharing one magnitude scale across many weights discards information about different weight magnitudes, and the scale values themselves require additional storage, adding overhead to the model size.
How does Boole-1.7B compare to other 1-bit models?: At nearly the same weight storage as binary Bonsai-1.7B, Boole averages 5.58 points higher (50.55 vs 44.97). Ternary Bonsai-1.7B scores 1.02 points higher but requires 1.9× more storage. Boole has the highest intelligence density among all evaluated models at 2.85 per GB.
What benchmark tasks were used to evaluate Boole-1.7B?: Evaluation uses an equal-weight mean of ten tasks: GSM8K, MATH-500, HumanEval+ and MBPP+, MMLU-Redux, GPQA-Diamond, MuSR, IFEval, IFBench, and BFCLv3. Answers are extracted by rule-based parsers with language-model fallback when parsing fails.
In which tasks does Boole-1.7B show the largest improvements over Bonsai?: Boole improves on binary Bonsai in seven of ten tasks, with the largest gains in math and code, while its scores are lower on MuSR, IFEval, and function calling (BFCLv3).

Find Boole-1.7B online

Leave a review for Boole-1.7B

Nobody has put Boole-1.7B on the bench yet. Leave the first review.

Reviewing is free, and nobody can pay for a review. Read the bench notes

Maker reviews · Workshop notes · Show it off · Did you make this? Claim it · Report this build

More builds like this

#1

Number 1: LAION

Open AI research for everyone
- Open source AI
- Germany
- DR 75 no change since last check
0 votes. Log in to give LAION a nod