LoginSign Up
← Back to all sparks
T

Together AI

AI-ASSISTANTS
Velocity5.0

Open-source AI cloud platform for training and inference of generative models

www.together.ai ↗

Together AI is building the operational stack around open models, not just the API.

open-modelscoding-agentspost-traininggpu-clustersenterprise-inference
◆Current state
Together shipped four infrastructure products in a month: Together Link for routing open models into existing coding agents, a dedicated B300 inference cluster with IBM Cloud and NVIDIA, Post-Training for custom models, and auto repair for GPU Clusters. Meta Muse Spark Open also went live with a 1M-token context window. The pitch has broadened from cheap inference to owning the full lifecycle of an open model.
◆Where it's heading
Together is moving up from inference provider to the operating layer for teams leaving closed models. Together Link addresses the switch, Post-Training the customization, and auto repair and dedicated capacity the production risk. Each removes a reason to stay on a closed API.
◆Prediction
Expect further tooling that shortens the path from closed to open models in coding agents, plus same-week support for major open releases. Pricing changes are not signaled in these entries.

◆Recent moves

  1. 18d ago

    Auto repair for Together GPU Clusters: from GPU fault to healthy node

    Auto repair detects GPU, host and fabric faults and applies the lightest fix, with per-fault control over what runs automatically. It reduces the operating burden of the clusters Together sells.

    View source ↗
  2. 26d ago

    Together Post-Training: frontier-grade model shaping, the way that fits you

    Post-Training lets teams train open models such as Qwen, GLM and Kimi into models they own, with SFT or RL, while Together runs the infrastructure. It extends the offer from serving models to shaping them.

    View source ↗
  3. 1mo ago

    Meta Muse Spark Open is live on Together AI

    Meta Muse Spark Open went live on serverless and dedicated infrastructure with a 1M-token context window. It keeps up the pattern of quick support for major open releases.

    View source ↗
  4. 1mo ago

    Expanding our enterprise inference capacity with IBM Cloud and NVIDIA

    A dedicated B300 inference cluster, built with IBM Cloud and NVIDIA, targets enterprise production workloads on open models. It adds capacity aimed at larger buyers.

    View source ↗
  5. 1mo ago

    Together Link: open models in the harness you already use. Start with one command today.

    Together Link brings models like GLM 5.3 and Kimi K3 into existing coding agents, and claims over 50% lower model spend. It turns the cost argument into a one-command switch.

    View source ↗
  6. 1mo ago

    How to train your own Jev for $17

    A tutorial on fine-tuning a small experimental classifier on Together's serverless platform. A demonstration rather than a product move.

    View source ↗