Together AI model pricing.

prices as of · 50 models · 51 prices

Serves open-weight models on its own GPUs; also sells dedicated endpoints by the GPU-hour.

Together AI is a host selling 50 models, from $0.24 per million tokens for Llama 3.2 1B Instruct, read on 16 Sept 2026.

Together AI prices

Kind of provider

Host

based in US

Where it runs

not stated

no regions published

How you pay

not stated

no payment terms published

Free tier

none published

nothing given away

Where Together AI's prices land, and why

Together AI publishes 51 prices we can compare, and 13 of them are below the list price for that model. 9 of those have a published reason behind them, so they rank like any other price. The other 2 sit below list on models whose weights are closed, with nothing published to account for the gap, so they are listed here in full and ranked nowhere on the site. Nobody here has read its terms yet, so treat what it says about itself as its own claim.

Together AI does not say whether it trains on what you send it. It does not say how long it keeps requests.

not yet checked: terms · privacy policy ↗ · status page ↗

Models to start with

all 51 prices ↓

Everything Together AI sells, against the list price

Cheapest first inside each group. The share is this row against the price the lab that trained the model charges.

Cheaper, and we know why

9 prices

Below list, with a published reason: hosting the open weights, a regional price list, credits, a stated discount.

  • DeepSeek V4 Flash 0731

    DeepSeek · 1M context

    7% under the list price, an ordinary reseller margin.

    $0.14 / $0.28in / out per M

    0.67×of list price

  • Qwen3-Next 80B-A3B Instruct

    Alibaba Cloud · 131k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $0.15 / $1.50in / out per M

    0.56×of list price

  • Qwen3-Next 80B-A3B (Thinking)

    Alibaba Cloud · 131k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $0.15 / $1.50in / out per M

    0.26×of list price

  • GLM-4.7

    Z.ai · 205k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $0.45 / $2in / out per M

    0.84×of list price

  • Kimi K2.5

    Moonshot AI · 262k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $0.50 / $2.80in / out per M

    0.90×of list price

  • Kimi K2.5

    Moonshot AI · 262k context

    Serves the open weights at fp4, a lower precision than the maker's, so not quite the same product.

    $0.50 / $2.80in / out per M

    0.90×of list price

  • Qwen2.5 72B Instruct

    Alibaba Cloud · 131k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $1.20 / $1.20in / out per M

    0.49×of list price

  • Inkling

    Thinking Machines · 1M context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $1 / $4.05in / out per M

    0.69×of list price

  • Qwen2.5-VL 72B Instruct

    Alibaba Cloud · 131k context

    Hosts the open weights on its own servers, so the price is its own, not a resale. Precision not disclosed.

    $1.95 / $8in / out per M

    0.82×of list price

Same price as the maker

40 prices

The same price the maker charges, sold by someone else.

showing the 30 cheapest of 40 · every row in the open dataset

Cheaper, and nobody says why

2 prices · never ranked

Below list on a closed model with nothing to account for it. Shown, never ranked, never a pick.

  • Qwen3.7 Plus

    Alibaba Cloud · 1M context

    0.64× Alibaba Cloud's list price on a closed model, and nobody says why. Listed, never ranked.

    $0.32 / $1.28in / out per M

    0.50×of list price

  • Qwen3.7 Max

    Alibaba Cloud · 1M context

    0.50× Alibaba Cloud's list price on a closed model, and nobody says why. Listed, never ranked.

    $1.25 / $3.75in / out per M

    0.50×of list price

Frequently asked

What is Together AI?
Serves open-weight models on its own GPUs; also sells dedicated endpoints by the GPU-hour. It sells 50 of the models we track, and trades as Together AI, based in US.
Is Together AI cheaper than buying from the maker?
13 of its 51 prices sit below the list price for the model, and 9 of those have a published reason behind them, so they can lead a ranking. 2 are listed and never ranked.
Does Together AI train on what you send it?
Together AI does not say whether it trains on what you send it. It does not say how long it keeps requests.
How current are these Together AI prices?
They come from the public catalogs we read, last read 16 Sept 2026. Prices exclude tax.