Vipul Ved Prakash says Together is #1 provider by token volume on OpenRouter
Together has reached the top ranking for token volume across closed and open endpoints on OpenRouter, according to Vipul Ved Prakash.
#together-ai
View all AI newsTogether AI partners with IBM and NVIDIA on dedicated B300 cluster for open-source inference
Together AI is scaling open-source model inference in collaboration with IBM and NVIDIA on a dedicated B300 cluster hosted on IBM Cloud.
Together AI says it leads OpenRouter token share for top open coding models
Together AI says it currently has the largest OpenRouter token share for GLM 5.3 Flash, DeepSeek V4.1 Flash, and Kimi K3, at 29.2%, 25.6%, and 18.9% respectively.
Together AI publishes Tev1-4B-experimental weights on Hugging Face
The Tev1-4B-experimental weights are in the togethercomputer repository on Hugging Face. Together AI describes the model as an experiment meant to show how specialized decision models can be trained fairly easily.
Together AI releases tev1-4B-experimental on Together serverless
The Jev-like classifier is finetuned on Qwen3.5 4B and priced at $0.042/M input and $0/M output. Together AI is also releasing a data recipe and a finetuning tutorial, and says training Tev1 cost $17.
Together AI offers 40% off Qwen3.7-Max and Qwen3.8-Flash on serverless through September 30
The discount applies to both Alibaba_Qwen models on Together AI serverless and runs through September 30. Together AI recommends Qwen3.7-Max for long-horizon agent work and Qwen3.8-Flash for high-volume, cost-sensitive workloads.
Self-organizing team of three models averages 66.7% in Stanford and Together AI paper
DAIR.AI reports that a Stanford and Together AI paper found o3-mini, Claude Sonnet 4, and DeepSeek-V3 averaged 66.7% across five math and physics benchmarks as a self-organizing team, above 48.8% for the strongest member alone and 59.0% for a perfect router.
Together AI adds canary rollouts to Dedicated Model Inference
Dedicated Model Inference can now upgrade the model behind a live endpoint without downtime. Traffic shifts in gated steps, and metric gates can pause a rollout if p95 latency or the error rate worsens.