Tiny tuned classifier beats Jev: GLiNER 2.5 hits 99.7% vs 83.6%, 8.8x faster locally
rickasaurus · x · 2026-09-19
Josh Kuechly argues that while Jev is strong at zero-shot classification, specialist classifiers will dominate commercial use cases. Evidence: trycua tuned a tiny model that scored 99.7% on their form-filling eval versus hosted Jev's 83.6%. The author tuned GLiNER 2.5 in just 51 minutes and it crushes Jev — locally and 8.8x faster, with a linked tutorial so anyone can replicate it. Reblogger rickasaurus adds that 'trainable Jev' will be a big deal once open-source copycats arrive.
More from Models
- Power user burns 1bn tokens a day at 98.65% cache hits, weighing GLM-5.3 vs DeepSeek after Claude Code weekly limits — julianharris · 2026-09-19
- Open models now outspend OpenAI on Vercel gateway, taking 78% of token volume — soumitrashukla9 · 2026-09-19
- Minimax-H3 dataset trending on Hugging Face under MIT license — stablediffusiontutorials · 2026-09-19
- Anthropic discloses four Claude alignment incidents; METR to investigate — TheZvi · 2026-09-19
- User's private benchmark finds Astra tiers degraded ~10x from release quality — PM__me_sth · 2026-09-19
- Four flagship models on the same sketch: unannounced Gemini 4 checkpoint spotted in Arena — CodeByPoonam · 2026-09-19