AIBespoke Nimble launches with an open model and synthetic data
The announcement reports 90% for Nimble on a curated evaluation, versus 66% for its Qwen base model and 93% for Jev. It cautions that Nimble could fare much worse than Jev on other benchmarks.
TLDR
Bespoke Nimble's release announcement introduces a fine-tuned Qwen3.5-9B model with open data and an open recipe. It describes fully synthetic data across 10 categories and a method called “contrastive data curation,” which slightly changes facts to generate negative examples. The team reports 90% on its curated evaluation, compared with 66% for Qwen and 93% for Jev. It cautions that there is no standard performance benchmark and that Nimble could perform much worse than Jev on other benchmarks.
Combined views
616
1 Source, first seen 3h ago
Bespoke Nimble launches with an open model and synthetic data
The announcement reports 90% for Nimble on a curated evaluation, versus 66% for its Qwen base model and 93% for Jev. It cautions that Nimble could fare much worse than Jev on other benchmarks.
TLDR
Bespoke Nimble's release announcement introduces a fine-tuned Qwen3.5-9B model with open data and an open recipe. It describes fully synthetic data across 10 categories and a method called “contrastive data curation,” which slightly changes facts to generate negative examples. The team reports 90% on its curated evaluation, compared with 66% for Qwen and 93% for Jev. It cautions that there is no standard performance benchmark and that Nimble could perform much worse than Jev on other benchmarks.