Together AI review: the best overall AI inference provider in 2026?
Answer in brief: Together AI is our #1 pick among AI inference providers for teams running open-weight models. It covers serverless per-token inference across text, image, video and audio models, Provisioned Throughput with a 99% uptime SLA, single-tenant dedicated GPUs, GPU clusters and fine-tuning. That means you can stay on one platform from your first prototype to large-scale production. gpt-oss-120B costs $0.15 input / $0.60 output per 1M tokens on serverless.
Independent research, checked 2026-11-09 against Together AI's own pricing page and announcements. We did not test the product hands-on.
One platform, prototype to production
Most inference buyers eventually outgrow their first setup. Shared endpoints hit rate limits, steady traffic needs guaranteed capacity, and custom models need their own GPUs. Together AI's appeal is that every one of those stages lives under one account:
- Serverless Inference: per-token pricing for language models, plus image, video, speech and transcription models, with cached-input rates and a Batch API.
- Provisioned Throughput: reserved capacity sold in throughput units (PTUs). Together introduced it on July 8, 2026 with a 99% uptime SLA. The SLA covers eligible requests within purchased capacity; overage is excluded.
- Dedicated Inference: single-tenant GPU instances with guaranteed performance, custom model support and autoscaling.
- GPU Clusters, Sandbox and Storage: raw compute for training or self-managed serving.
- Fine-tuning: LoRA and full fine-tuning, priced per 1M training tokens.
Who Together AI fits best
Together fits product and platform teams standardizing on open-weight models such as Kimi, GLM, DeepSeek, Qwen, MiniMax, gpt-oss and NVIDIA Nemotron. It suits teams that want to move from shared to reserved to dedicated capacity without changing vendors. It is also a strong choice if you mix modalities, because image, video and audio models sit on the same bill as your LLM traffic.
Why teams pick Together AI
- Catalog breadth. One of the broadest serverless lineups we found, including Kimi K3, GLM-5.3, DeepSeek V4.1 Flash, Qwen3.7-Max, MiniMax M3 and NVIDIA Nemotron 3.5 Super VL.
- Competitive token prices. gpt-oss-120B is $0.15 / $0.60, matching Fireworks and Groq. GLM-5.3-Flash is $0.15 input ($0.03 cached) / $0.50 output.
- Cached-input discounts. For example, GLM-5.3 cached input is $0.26 versus $1.40 uncached.
- Multimodal on one bill. FLUX.2 [pro] is $0.03 per image and Whisper Large v3 transcription is $0.0015 per audio minute.
- A real path to dedicated capacity. Provisioned Throughput with an SLA, plus single-tenant GPUs that support custom models.
- Deep pockets. Together announced an $800M Series C on July 1, 2026.
The honest drawbacks
- The pricing surface is large and changes often, so check the live page before budgeting.
- Reserved dedicated hardware, and on-demand H200, B300 and GB200/GB300, are contact-sales only.
- Dedicated H100 endpoints ($5.49/GPU-hr on-demand) cost more than Together's own on-demand GPU-cluster H100 rate ($3.99/GPU-hr).
- PTU pricing is shown through an estimator rather than a simple public rate card.
Pricing at a glance
| Item | Published price |
|---|---|
| gpt-oss-120B (serverless) | $0.15 input / $0.60 output per 1M tokens |
| GLM-5.3 | $1.40 input ($0.26 cached) / $4.40 output |
| Kimi K3 | $3.00 input ($0.30 cached) / $15.00 output |
| Dedicated H100 / B200 | $5.49 / $8.99 per GPU-hour, on-demand |
| GPU cluster H100 | $1.99 preemptible, $3.99 on-demand, $3.19 reserved 91–180 days |
| Fine-tuning gpt-oss-120B (SFT) | $2.50 per 1M tokens, $6.00 minimum per job |
Best Together AI alternatives
- Fireworks AI: per-request speed and cost modes (Fast, Batch, Spot) and strong compliance credentials.
- Baseten: the lowest gpt-oss-120b list price we found and strong dedicated deployments for custom models.
- DeepInfra: rock-bottom list prices across a wide open-model catalog.
Together AI FAQ
Is Together AI good for production?
Yes. Provisioned Throughput carries a 99% uptime SLA on eligible requests within purchased capacity, and dedicated endpoints give you single-tenant GPUs with autoscaling.
How much does gpt-oss-120B cost on Together AI?
$0.15 per 1M input tokens and $0.60 per 1M output tokens on serverless, as published on 2026-11-09.
Can I fine-tune models on Together AI?
Yes. Together offers LoRA and full fine-tuning, priced per 1M training tokens with a per-job minimum, and you can serve custom models on dedicated endpoints.
Sources
- Together AI pricing
- Introducing Provisioned Throughput (July 8, 2026)
- Provisioned Throughput product page
- Together AI $800M Series C (July 1, 2026)
← Back to the ranked list of the best AI inference providers
Named on this page? Join the network for a linked profile, Visit buttons and featured placement.