Sign in Subscribe

Benchmark

Dark thumbnail: headline Don't trust our benchmark. Audit it. over a memory-bank motif with two glowing result records; pills show 98.4% LongMemEval-S, 45ms recall, 0 LLM tokens.

Don’t Trust Our Benchmark. Audit It.

107 of 108 identical. Benchmark reproducibility thumbnail with two overlapping waveforms merging into one bright point

We published our benchmark. Then we tried to break it.

Dark abstract purple vector ribbons on black with the headline: 87GB on a CPU. Quantization did that.

An 87GB model on a CPU: what the quantization math buys you

Two cyan light streams converging into one bright point on a black background, representing fusion recall

Uteke v0.16.0: Better Recall With Zero Configuration

Covecto vs Potrace vs vtracer: SVG Quality Compared

Covecto vs Potrace vs vtracer: SVG Quality Compared

Cora Code Performance: Benchmarking 50K-Line Codebases

Cora Code Performance: Benchmarking 50K-Line Codebases