Together AI released tev1-4B-experimental, a small classifier model fine-tuned on top of Alibaba’s Qwen3.5 4B, in a thread posted on X. The company priced the model at $0.042 per million input tokens on its serverless platform, with output tokens free.
Together AI said the entire training run cost $17, a figure the company reported itself and did not independently verify. Alongside the model, the company published its data recipe and a tutorial walking developers through fine-tuning a custom classifier on the same base.
Classifiers built this way are typically used for tasks like content moderation, routing, and filtering rather than open-ended generation, which is why the training cost stays low compared with frontier language models. The zero-cost output pricing signals Together AI is angling to win high-volume classification workloads away from larger, more expensive general-purpose models.
Teams currently paying per-token for classification tasks on frontier models should benchmark tev1-4B-experimental’s accuracy before assuming the cheaper price holds up in production.
Together AI announced tev1-4B-experimental in a thread on X on September 24, 2026.
