Skip to content

Benchmark

BABILong avg(qa1-5) 64k

Long context benchmarks measure how well models use information in long inputs. Scores are percentages. Higher is better.

Models and systems measured
12
Scores recorded
12
Sources
1
IQ weight
No weight published

Score distribution

53.1%
Mean
48%
Median
22.7%
Std dev
35%
10th percentile
97%
90th percentile
21%
Min
21%97.9%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
~ ARMT (137M) fine-tuneavg, 64k· BABILong v0 data docs97.9%
~ Mamba (130M) fine-tuneavg, 64k· BABILong v0 data docs97%
~ RMT (137M) fine-tuneavg, 64k· BABILong v0 data docs69.9%
ChatGLM3 6Bchatglm3-6b-128k, avg, 64k, 128k· BABILong v0 data docs21%
GPT-4-0125-previewavg, 64k· BABILong v0 data docs43%
Llama 3.1 70B InstructMeta-Llama-3.1-70B-Instruct, avg, 64k· BABILong v0 data docs53%
Llama 3.1 8B InstructMeta-Llama-3.1-8B-Instruct, avg, 64k· BABILong v0 data docs49%
Llama3-ChatQA-1.5-8B + RAGavg, 64k· BABILong v0 data docs42%
Mixtral 8x22B Instruct v0.1Mixtral-8x22B-Instruct-v0.1, avg, 64k· BABILong v0 data docs35%
Phi-3 Medium 128K InstructPhi-3-medium-128k-instruct, avg, 64k· BABILong v0 data docs45%
Phi-3 Mini 128K InstructPhi-3-mini-128k-instruct, avg, 64k· BABILong v0 data docs37%
Yi-34B01-ai/Yi-34B-200k, avg, 64k, 200k· BABILong v0 data docs48%

Data credits: RMT-team, Kuratov et al. Official v0 numeric CSV; results licence not separately stated.