Dedicated Model Inference

Deploy models on dedicated infrastructure, engineered for speed

Purpose-built for teams who need control and the best economics in the market.

Why Dedicated Inference with Together AI?

Designed for production workloads that need consistent performance and operational control.

Built for production inference

Scale to thousands of GPUs for always-on, production inference deployments.

Industry-leading unit economics

We provide the fastest deployments, enabling best price-performance on top GPUs.

Powered by frontier AI systems research

We continuously roll out the latest innovations to keep your deployments running fast.

Research that ships

Our research team doesn't just publish. They build the optimizations that power every inference request.

  • ATLAS
  • CPD
  • Megakernel
  • ThunderKittens
  • Performance on DeepSeek V3.1 (Arena Hard)

    • Atlas
    • Static Speculator
    • No Speculator

    ATLAS performance

    3.18x faster

    ATLAS, our AdapTive-LeArning Speculator System, continuously learns from live traffic — outperforming static speculators and specialized hardware.

    learn more
  • CPD improves sustainable QPS by 35-40%

    • CPD
    • Baseline

    Together AI CPD vs 2P1D

    +40% throughput

    Long-context inference without the latency penalty. CPD (cache-aware prefill-decode disaggregation) separates warm and cold requests, cutting time-to-first-token and boosting throughput by up to 40%.

    learn more
  • Time to first 64 tokens

    • Megakernel (H100)
    • Baseline (B200)

    Megakernel vs baseline

    Up to 3.6x faster

    Megakernel fuses an entire model's forward pass into a single GPU kernel. Made using the ThunderKittens framework, Megakernel eliminates the idle gaps between operations that rob GPUs of their full potential.

    learn more
  • BF16 all-reduce sum performance (on 8x NVIDIA B200s)

    • PK
    • NCCL

    ParallelKittens vs NCCL

    Up to 1.79x faster

    ParallelKittens—an extension to ThunderKittens for multi-GPU workloads developed in collaboration with Stanford's Hazy Lab—cuts the synchronization overhead that large multi-GPU models pay on every single forward pass.

    learn more

Build with leading models

Explore top-performing models across text, image, video, code, and voice.

Chat

GLM-5.3

new

Chat

DeepSeek V4.1 Flash 1017

new

Chat

Kimi K3.1 Pro

new

Chat

Inkling 2 Small

new

Chat

Qwen3.8 Flash

new

Chat

Muse Spark Open 816B A42B

new

Chat

Beam 1.0 501B A23B

new

Chat

MiniMax M3.1 Flash

new

Audio

Qwen3 TTS 12Hz 1.7B

new

Chat

Qwen4 Max

new

Chat

Tencent Hy4

Video

ByteDance Seedance 2.5

new

Chat

GLM-5.3-Flash

new

Video

MiniMax H3

new

Chat

NVIDIA Nemotron 3.5 Super VL

Chat

Gemma 4 31B

new

Chat

Prism ML Ternary Bonsai 2 27B

new

Chat

Elorian Glimpse 1.0

new

Chat

Ripple-65B Preview

Chat

DeepSeek V4 Pro