Together AI review
Updated September 28, 2026 · DeployCue Editorial Team · Facts checked September 26, 2026
We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings.
Our verdict
Together AI suits developers who want a wide model catalogue and dedicated NVIDIA GPUs from a single vendor. Input price sits at 1 of 71, though there is no free trial, so budget a small prepaid balance before your first call and check the per-model rates before committing.
Key facts
- Best for
- Single-vendor AI workloads where entry price and a broad model list matter more than very long context.
- Input price from
- $0 to $3 (Source, checked September 26, 2026)
- Output price from
- $0 to $15 (Source, checked September 26, 2026)
- Models offered
- Inkling, MiniMax M3, Qwen3.8-2.4T-A95B, Qwen3.7 Max, Qwen3.6 Plus, Qwen3.5 9B, Kimi K3, GLM-5.3, GLM-5.3 Flash, GLM-5.2, GPT-OSS 120B, DeepSeek V4 Pro 0813, DeepSeek V4.1 Flash, DeepSeek-V4-Flash-0731, Llama 3.3 70B Instruct Turbo, Qwen3.7 Plus, Qwen3.8 Flash, Muse Glimmer 30B, Ternary Bonsai 27B, Tev1 4B Experimental, GPT Image 2, GPT Image 1.5, FLUX.2 dev, FLUX.2 pro, FLUX.2 max, Imagen 4.0 Preview, Sora 2, Veo 3.1, Kling 2.1 Pro, Seedance 2.5, Whisper Large v3, NVIDIA Parakeet TDT 0.6B v3, Kokoro-82M, Cartesia Sonic 3 (Source, checked September 26, 2026)
- Largest context window
- 1,048,576 (Source, checked September 26, 2026)
- Free credits
- No free trial; access requires a minimum $5 prepaid credit purchase (Source, checked September 26, 2026)
- OpenAI-compatible API
- Yes (Source, checked September 26, 2026)
- Legal entity
- Together Computer, Inc. (Source, checked September 26, 2026)
- Headquarters
- United States (Source, checked September 26, 2026)
- Founded
- 2022 (Source, checked September 26, 2026)
- Support channels
- Help-centre, Community (Source, checked September 26, 2026)
- Official website
- together.ai
GPU cloud
- GPU models
- NVIDIA HGX H100, NVIDIA HGX H200, NVIDIA HGX B200, NVIDIA HGX B300, NVIDIA GB200 NVL72, NVIDIA GB300 NVL72 (Source, checked September 26, 2026)
- GPU hour from
- $1.99 (Source, checked September 26, 2026)
- Billing granularity
- Hourly (Source, checked September 26, 2026)
- Egress fee
- No data ingress or egress fees (Source, checked September 26, 2026)
- Free credits
- No free trial; access requires a minimum $5 prepaid credit purchase (Source, checked September 26, 2026)
- Legal entity
- Together Computer, Inc. (Source, checked September 26, 2026)
- Headquarters
- United States (Source, checked September 26, 2026)
- Founded
- 2022 (Source, checked September 26, 2026)
- Support channels
- Help-centre, Community (Source, checked September 26, 2026)
- Pricing model
- Hourly billing, Free tier
Apps and profiles
Pricing
Together AI gpt-oss-120b (USD)
input-tokens: $0.15 / million
Together AI gpt-oss-120b (USD)
output-tokens: $0.60 / million
Together AI Mixtral 8x7B (USD)
input-tokens: $0.90 / million
Together AI Mixtral 8x7B (USD)
output-tokens: $0.90 / million
Together AI Mixtral 8x22B (USD)
input-tokens: $2.40 / million
Together AI Mixtral 8x22B (USD)
output-tokens: $2.40 / million
Together AI Llama 3.3 70B (USD)
input-tokens: $0.88 / million
Together AI Llama 3.3 70B (USD)
output-tokens: $0.88 / million
Pros and cons
Pros
- Ranks 1 of 71 providers on input price and 1 of 70 on output price, so entry budgets stretch further here.
- Data transfer is straightforward: No data ingress or egress fees, so cluster jobs do not pick up a separate transfer line.
- Dedicated NVIDIA HGX H100, NVIDIA HGX H200 and NVIDIA HGX B200 sit behind the same Together AI account as the API, with GPU time at $1.99 an hour.
- Inkling, MiniMax M3 and Qwen3.8 2.4T A95B covers text, image, video and speech, so a media pipeline can stay in one place.
- The OpenAI-compatible API is a yes, so existing clients and request bodies carry over with small edits.
Cons
- A free trial is not among the billing options listed as hourly and free tier, and access requires a minimum prepaid credit purchase, so any test run starts with a balance already funded.
- Support runs on help centre and community, so there is no live channel to lean on during an incident.
- A largest context window of 1,048,576 places the brand at 11 of 72, the weaker side of the sheet.
- GPU time starts at $1.99 where the group median is €1.40, and hourly billing suits steady runs more than quick tests.
Who it suits
- Developers who want chat, image, video and speech models behind a single OpenAI-compatible API rather than several vendor accounts.
- Teams already committed to NVIDIA HGX H100, NVIDIA HGX H200 and NVIDIA HGX B200 who want inference and training work on the same platform.
- Buyers shopping mainly on entry price, since input price ranks 1 of 71 providers.
Who should look elsewhere
- Teams that need a named human contact on contract: Azure AI Foundry is the platform to compare, while support here runs on help centre and community.
- Developers who want Claude, Gemini or Mistral: Supascale AI lists DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra.
- Buyers who need a no-cost run before spending: Salad, with input price from $0.06, is the alternative to price against, while access here needs a minimum prepaid credit purchase.
How it compares on our data
On our numbers, Together AI comes out well on entry price and mid-table on long context. Input price ranks 1 of 71 providers, and output price ranks 1 of 70, so the spend-per-request maths is as kind as it gets in this set. The largest context window is the other side of the ledger, sitting at 11 of 72. If your budget is driven by volume of requests, this is a strong sheet. If it is driven by very long documents or long agent histories, read the context figure before you plan around it.
Core features
The core of the platform is a single catalogue listed as Inkling, MiniMax M3 and Qwen3.8 2.4T A95B, spanning chat models, image generators, video models and speech tools. That mix matters if you would rather not stitch several vendors together: a text pipeline and a thumbnail job can share the same account. The OpenAI-compatible API is a yes, so request bodies, streaming calls and client libraries carry across from OpenAI tooling with small edits. Headline limits are good too, with a largest context window of 1,048,576 tokens, so long documents fit in a request rather than being chunked by hand.
Plans, egress and billing
Together AI prices sit across a wide band, with input price running $0 to $3 and output price $0 to $15. The low end of each range is the practical entry point for a first workload, while the high end is what the premium models cost, so the same platform covers a cheap pilot and a production model. The figure to budget with is the one attached to the model you actually plan to call. Payment is prepaid: there is no free trial, and access requires a minimum credit purchase, so top up a little before your first request rather than meeting a balance requirement mid-run.
Models and context
The catalogue in Inkling, MiniMax M3 and Qwen3.8 2.4T A95B covers open-weight chat models alongside image, video and speech generation, so most jobs can be served from the same vendor. The largest context window is 1,048,576 tokens, which is what makes long documents, long transcripts and long tool-calling traces practical without manual chunking. Smaller models in the same list trade context for price, so context is a per-model choice rather than a platform-wide limit. Because the OpenAI-compatible API is a yes, moving between those models is a parameter change rather than a rewrite.
The company behind it
Together AI is operated as Together Computer, Inc., headquartered in the United States and founded in 2022. That is a young company by the standards of the long-established cloud platforms, so the sensible question for a team is how stable the catalogue looks over your planning horizon. Support runs through help centre and community, which suits engineers who are comfortable reading documentation and asking peers in a forum. If you need a named contact while a contract is being drawn up, raise that early rather than after spend has started.
How Together AI compares
Together AI vs BytePlus9.4 vs 8.7
Together AI vs RunPod9.4 vs 9.5
Together AI vs Zhipu AI (Z.ai)9.4 vs 9.4
Together AI vs WaveSpeedAI9.4 vs 8.8
Together AI vs AIS Cloud9.4 vs 8.1
Frequently asked questions
What does it cost to run a model on Together AI?+
Input price runs $0 to $3 and output price $0 to $15 across the catalogue. The range is wide because it covers everything from cheap open-weight models to premium ones, so the figure that matters for you is the one attached to the specific model you plan to call. Budgets are prepaid rather than billed on a monthly cycle.
Can I try Together AI before paying anything?+
There is no free trial, and access requires a minimum prepaid credit purchase, so a balance needs to exist before your first request. Billing still offers a free tier alongside hourly charges, shown as hourly and free tier, so free-tier models stay reachable once that balance is in place.
How much is a GPU cluster per hour?+
GPU time starts at $1.99 on NVIDIA HGX H100, NVIDIA HGX H200 and NVIDIA HGX B200, and clusters bill at hourly granularity. For comparison, the median GPU hour across this group is €1.40, so check the price of the exact card you need before committing to a long run.
Are there egress or data transfer fees?+
Data transfer is straightforward: No data ingress or egress fees. That matters for fine-tuning work that pulls datasets in and pushes checkpoints out, because the cost stays in the $1.99 GPU time rather than appearing as a separate line on the same account.
Which models can I call, and how much context do I get?+
The catalogue in Inkling, MiniMax M3 and Qwen3.8 2.4T A95B spans chat, image, video and speech models from several families. A largest context window of 1,048,576 tokens is available on the largest models, which is what makes long documents and long transcripts workable without hand-splitting them first.
Can I reuse my existing OpenAI code?+
The OpenAI-compatible API is a yes, so client libraries, request bodies and streaming calls carry over with small edits. The catalogue behind that endpoint is Inkling, MiniMax M3 and Qwen3.8 2.4T A95B, so a migration usually means changing the model name rather than rewriting the integration layer.
Who runs Together AI?+
The service is operated by Together Computer, Inc., headquartered in the United States and founded in 2022. That is the entity to name when your procurement team asks who the counterparty is, and the figure to weigh if vendor longevity matters to your plan.
How do I get help when something breaks?+
Support runs on help centre and community. In practice that means documentation you read yourself plus a community forum, so the speed of an answer depends on how searchable the problem is. If your team depends on a live channel during an incident, settle that question before moving production traffic.
Sources
- docs.together.ai/docs/billing checked September 26, 2026
Backs: Free credits
“Together AI does not currently offer free trials. Access to the Together platform requires a minimum $5 credit purchase.”
- docs.together.ai/docs/gpu-clusters-billing.md checked September 26, 2026
Backs: Egress fee
“No, Together AI does not charge for data transfer into or out of your cluster”
- docs.together.ai/docs/openai-api-compatibility checked September 26, 2026
Backs: OpenAI-compatible API, Support channels
“Together's API is compatible with the OpenAI REST API and SDKs across chat, completions, vision, image generation, text-to-speech, and embeddings.”
- docs.together.ai/docs/serverless/models checked September 26, 2026
Backs: Largest context window, Models offered
“Moonshot | Kimi K3 | moonshotai/Kimi-K3 | 1048576 | $3.00 | $0.30 | $15.00”
- together.ai checked September 26, 2026
Backs: Founded
“"@type": "Organization", "name": "Together AI", "foundingDate": "2022"”
- together.ai/pricing checked September 26, 2026
Backs: Billing granularity, GPU hour from, GPU models, Input price from, Pricing model, Output price from
“All prices are per GPU per hour.”
- together.ai/terms-of-service checked September 26, 2026
Backs: Headquarters, Legal entity
“Together Computer, Inc. Address: 251 Rhode Island Street, Suite 200, San Francisco, CA 94103”
Compiled from 7 source pages, checked September 26, 2026. We did not open an account.
Together AI
Single-vendor AI workloads where entry price and a broad model list matter more than very long context.