Together releases Tev1-4B, a Jev-like classifier fine-tuned on Qwen3.5 4B, trained for just $17
RexDouglass · x · 2026-09-24
Together launched the experimental tev1-4B-experimental classifier, fine-tuned from Qwen3.5 4B with a Jev-like approach, available on Together serverless at $0.042/M input and $0/M output. The data recipe and a fine-tuning tutorial are also public — Tev1 reportedly cost only $17 to train.
N8Programs cautions that a Jev-like classifier label doesn't imply weak capability: it dominates on a wide range of benchmarks, sometimes by 20 percentage points.
More from Models
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12
- Gemini re-renders images as text when vision is removed, dropping only 6% WSR — zacharynado · 2026-11-12
- Bare-harness coding benchmark: Fable 5.2 best overall, Gemini 4 Carbon best value — Semantic_meaning · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Reddit's 2026 LLM hall of fame: Opus 5.5 still unbeaten after 9 rival releases — G0dZylla · 2026-11-12
- Commit leak reveals unreleased Gemini 4 variants: Argon, Hydrogen and Helix (unconfirmed) — BlackHC · 2026-11-11