Enki

Together AI review: what its API and GPUs cost

Inference, fine-tuning and GPU clusters for open models

7.9Editor’s scorePaidfrom Pay as you goPricing verified Nov 17, 2026

Screenshots

Together AI

Serverless inference and GPU clusters for open models

7.9OUT OF 10

Enki editor score

Great

Human-vetted editorial rating

Earned by low serverless rates such as gpt-oss-120B at $0.15/$0.60 and H100s from $1.99 an hour; cost by no free trial, a $15 minimum first purchase and pricing spread across many products.

We may earn a commission if you sign up through our links. It never affects our ratings or which stories we cover.

Overview

Together AI sells the infrastructure behind open-model apps: serverless APIs priced per million tokens for models such as gpt-oss, Qwen, DeepSeek, GLM and Kimi, dedicated endpoints and GPU clusters billed per GPU-hour, and fine-tuning priced per million training tokens. It suits developers who have outgrown a single hosted API and want to fine-tune or reserve their own capacity without leaving one provider. The account is prepaid: the first credit purchase is at least $15, then $5, and Together says it does not currently offer free trials. Scale and Enterprise contracts add committed throughput.

Enki verdict

Together AI is best at covering the whole open-model stack, from cheap per-token inference to fine-tuning and reserved GPU clusters, from one prepaid account. Reach for it when you are past prototyping and want to fine-tune or own capacity for open models, not for a founder who only needs a free API key to try a model.

Key features

Serverless APIs

Per-token access to open models such as gpt-oss, Qwen, DeepSeek and Kimi.

Fine-tuning

SFT and DPO training priced per million tokens.

GPU clusters

H100 and B200 capacity, preemptible, on demand or reserved.

Pros & cons

What we like

  • Inference, fine-tuning and GPU clusters on one account
  • Low serverless rates on open models
  • Preemptible H100s from $1.99 an hour

Worth noting

  • No free trial
  • $15 minimum first purchase
  • Prices vary widely by model and product

Together AI pricing

Together AI bills each product by usage from a prepaid balance: tokens for serverless, GPU-hours for dedicated endpoints and clusters.

Is Together AI free?

No. Together AI says it does not currently offer free trials, and the account is prepaid with a $15 minimum first purchase.

  • First credit purchase is at least $15, then at least $5
  • One serverless model, Ternary Bonsai 27B, is listed at $0 input and $0 output
Together AI plans and prices
Serverless inferencePer 1M tokensdeducted from prepaid credits
  • gpt-oss-120B: $0.15 input, $0.60 output
  • DeepSeek V4.1 Flash: $0.30 input, $1.20 output
  • Kimi K3: $3.00 input, $15.00 output
Dedicated inferenceFrom $5.49/GPU-hron demand; reserved pricing through sales
  • NVIDIA HGX H100 $5.49 per GPU-hour
  • NVIDIA HGX B200 $9.29 per GPU-hour
GPU clustersFrom $1.99/GPU-hrpreemptible, on demand or reserved for 3 to 90 days
  • H100: $1.99 preemptible, $3.99 on demand, $3.45 reserved 31 to 90 days
  • B200: $4.65 preemptible, $9.29 on demand
Fine-tuningPer 1M training tokensdataset size times epochs, with a per-model minimum charge
  • Llama 3.3 70B SFT from $2.03, minimum $4.00
  • gpt-oss-120B SFT $2.50, minimum $6.00

Source: www.together.ai/pricing
Checked on: . Prices can change; confirm at checkout.

What it means for you

Start with serverless inference and the $15 minimum top-up; it is the cheapest way to test open models at production quality. Move to dedicated inference or GPU clusters only when steady traffic or fine-tuning justifies paying by the hour, and use preemptible H100s for training jobs that can restart. If you want one key across open and closed models, OpenRouter routes to more providers, and Ollama runs small open models locally for free. See the best AI coding tools.

Price history

  1. Checked against https://www.together.ai/pricing: gpt-oss-120B $0.15/$0.60 per 1M tokens; dedicated H100 $5.49 and B200 $9.29 per GPU-hour; H100 clusters $1.99 preemptible to $3.99 on demand. Billing docs: no free trials, $15 minimum first purchase.

What real users think

  • “I regularly use together.ai with models like this, where they're fully self hosting them and I can turn off all data retention, at a fraction of the cost of proprietary models.”

  • “When using together.ai with Kilo Code the LLM was having a lot of trouble making successful edits. … Meanwhile it used the tools from my harness just fine.”

  • “It's a scam. I wanted to do an experimentation on training nano LLM. Then they charged me 160USD per month and impossible to cancel it. Still no reply from the support team.”

Quoted word for word from public posts, shortened only where marked with “…”. We pick a spread of views, not the kindest ones, and these comments do not affect the Enki score. Checked Nov 18, 2026.

Together AI FAQ

Is Together AI free?
No. Together AI's billing docs say it does not currently offer free trials, and the account is prepaid, with a minimum first purchase of $15. Its pricing page does list one serverless model, Ternary Bonsai 27B, at $0.
How much does Together AI cost?
Serverless models are priced per million tokens, for example $0.15 input and $0.60 output for gpt-oss-120B, or $3.00 and $15.00 for Kimi K3. Dedicated H100 endpoints are $5.49 per GPU-hour, and H100 cluster GPUs run from $1.99 an hour preemptible to $3.99 on demand.
Can I fine-tune models on Together AI?
Yes. Fine-tuning is billed per million training tokens, based on dataset size times epochs, with a minimum charge per model, such as $4.00 for Llama 3.1 8B.
Is Together AI prepaid?
Yes. Together says it is fully prepaid: you buy credits and usage is deducted. Scale and Enterprise customers with contracts are billed under their contract terms.

No written reviews yet — be the first to share your experience.

Related tools