AI API price index · open-model host · read 16 Oct 2026
Together AI API pricing
51 prices as Together AI publishes them — 28 text models, 5 speech to text, 12 images and documents, 6 hardware time — read on its own pricing page on 16 Oct 2026. Its current text models cost $0.14 to $15.00 per million output tokens.
Free tier Pricing page only says 'Start for free, scale on demand' (page meta description); no credit amount. The official Credits doc says: 'Together AI does not currently offer free trials. Access to the Together platform requires a minimum $15 credit purchase.' Two chat models are priced at zero on the pricing page: Ternary Bonsai 27B ($0.00 / $0.00) and Tev1 4B Experimental (output 'Free').
What to know before you build on Together AI
- There is no free trial: the first purchase is at least $15 of credit.
- Kimi K3's launch promotion ($2.70 in, $13.50 out) ended on 11 October 2026; it is $3.00 in, $0.30 cached and $15.00 out (re-read 16 October).
- The pricing page and Together's own docs disagreed on four models on the day we read them — Qwen3.7-Max, for one, was $1.50 / $4.50 on the page and $2.50 / $7.50 in the docs. The index uses the pricing page.
- Batch is "up to 50% off", but the only model listed at 50% is Whisper Large v3.
- No embedding model is offered serverless.
Against the other ten
Across the index, 85 current text models run from $0.10 to $50.00 per million output tokens — 500×. Together AI’s cheapest, Llama 3 8B Instruct Lite at $0.14, has 1 cheaper text model elsewhere; its dearest, Kimi K3, costs $15.00. A cheaper model is a different model, not the same one for less — compare capability before price.
Speech to text across the index runs from $0.04 to $0.552 per audio hour; Together AI’s NVIDIA Nemotron 3 ASR Streaming 0.6B is $0.09.
The same model elsewhere
Open-weight models Together AI sells that other hosts price too — the only like-for-like comparison in the index.
- Whisper Large v3 — Together AI $0.09/h · Groq $0.111/h — 1.2× apart
- DeepSeek V4.1 Flash — Fireworks AI $0.30 / $1.20 · Together AI $0.30 / $1.20 — the same everywhere
- GLM 5.3 — Fireworks AI $1.40 / $4.40 · Mistral AI $1.40 / $4.40 · Together AI $1.40 / $4.40 — the same everywhere
- GLM 5.3 Flash — Fireworks AI $0.15 / $0.50 · Together AI $0.15 / $0.50 — the same everywhere
- gpt-oss-120b — Fireworks AI $0.15 / $0.60 · Groq $0.15 / $0.60 · Together AI $0.15 / $0.60 — the same everywhere
- Kimi K3 — Fireworks AI $3.00 / $15.00 · Together AI $3.00 / $15.00 — the same everywhere
- MiniMax M3 — Fireworks AI $0.30 / $1.20 · Together AI $0.30 / $1.20 — the same everywhere
Input / output per 1M tokens, standard tier.
Text models
| Model | Input | Cached input | Output | Note |
|---|---|---|---|---|
| MiniMax M3serverless | $0.30 | $0.06 | $1.20 | — |
| Kimi K3serverless | $3.00 | $0.30 | $15.00 | List price. Re-read 16 Oct: the promotion the 10 Oct page printed ('Promotion valid until 10/11/2026', $2.70 / $0.27 cached / $13.50) has ended; the page and the docs model table both print $3.00 / $0.30 cached / $15.00. |
| Qwen3.8-2.4T-A95Bserverless | $2.00 | $0.25 | $6.00 | Docs model table prints cached input $0.50 |
| Muse Glimmer 30Bserverless | $0.35 | $0.04 | $1.50 | — |
| DeepSeek V4 Pro 0813serverless | $1.32 | $0.13 | $3.96 | — |
| DeepSeek V4.1 Flashserverless | $0.30 | $0.006 | $1.20 | — |
| DeepSeek V4 Flash 0731serverless | $0.14 | $0.03 | $0.28 | — |
| GLM-5.3serverless | $1.40 | $0.26 | $4.40 | — |
| GLM-5.3-Flashserverless | $0.15 | $0.03 | $0.50 | — |
| GLM-5.2serverless | $1.40 | $0.26 | $4.40 | — |
| gpt-oss-120Bserverless | $0.15 | — | $0.60 | — |
| Llama 3.3 70Bserverless | $1.04 | — | $1.04 | — |
| Llama 3 8B Instruct Liteserverless | $0.14 | — | $0.14 | — |
| Qwen3 235B A22B Instruct 2507 FP8 Throughputserverless | $0.20 | — | $0.60 | — |
| Qwen3.5-397B-A17Bserverless | $0.60 | $0.35 | $3.60 | — |
| Qwen3.5 9Bserverless | $0.17 | — | $0.25 | — |
| Qwen3.6-Plusserverless | $0.50 | — | $3.00 | — |
| Qwen3.7-Plusserverless | $0.32 | — | $1.28 | — |
| Qwen3.7-Maxserverless | $1.50 | $0.30 | $4.50 | Docs model table prints $2.50 / $0.50 cached / $7.50 for 'Qwen3.7 Max' — conflict with the pricing page |
| Qwen3.8 Flashserverless | $0.15 | — | $0.47 | — |
| Qwen2.5 7B Instruct Turboserverless | $0.30 | — | $0.30 | — |
| MiniMax M2.7serverless | $0.30 | $0.06 | $1.20 | — |
| Gemma 4 31Bserverless | $0.39 | — | $0.97 | — |
| Inklingserverless | $1.00 | $0.17 | $4.05 | — |
| Cogito v2.1 671Bserverless | $1.25 | — | $1.25 | — |
| Rnj-1 Instructserverless | $0.15 | — | $0.15 | — |
| Ternary Bonsai 27Bserverless | $0.00 | — | $0.00 | Priced at zero (docs print 'Free') · Not ranked: printed as free — no per-token price to rank. |
| Tev1 4B Experimentalserverless | $0.04 | — | $0.00 | Output printed as 'Free'; docs print input $0.042 · Not ranked: printed as free — no per-token price to rank. |
Speech to text
| Model | Price | Note |
|---|---|---|
| Whisper Large v3serverless | $0.0015 per minute | Per audio minute; docs list openai/whisper-large-v3 as 50% off in the Batch API |
| Whisper Large v3 (Streaming)serverless | $0.0035 per minute | Per audio minute |
| NVIDIA Parakeet TDT 0.6B v3serverless | $0.0015 per minute | Per audio minute |
| NVIDIA Nemotron 3 ASR Streaming 0.6Bserverless | $0.0015 per minute | Per audio minute |
| NVIDIA Nemotron 3.5 ASRserverless | $0.0045 per minute | Per audio minute (docs list 'NVIDIA Nemotron 3.5 ASR Streaming 0.6B' at $0.0015 — different model string) |
Images and documents
| Model | Price | Note |
|---|---|---|
| GPT Image 2serverless | $0.053 per image | Per image; docs: per-image prices are estimates and vary with resolution/quality |
| GPT Image 1.5serverless | $0.034 per image | Per image; estimate, varies with parameters |
| Nano Banana Pro (Gemini 3 Pro Image)serverless | $0.134 per image | Per image at 1K or 2K; docs: about $0.24 at 4K |
| Gemini 3.1 Flash Image (Nano Banana 2)serverless | $0.05 per image | Per image; docs print $0.04657 at 512x512, varies |
| Gemini Flash Image 2.5 (Nano Banana)serverless | $0.039 per image | Per image; estimate, varies |
| FLUX.2 [pro]serverless | $0.03 per image | Per image; estimate, varies |
| FLUX.2 [dev]serverless | $0.0154 per image | Per image; estimate, varies |
| FLUX.2 [flex]serverless | $0.03 per image | Per image; estimate, varies |
| Ideogram 4.0serverless | $0.06 per image | Per image; estimate, varies |
| Qwen Image 2.0serverless | $0.04 per image | Per image; docs print $0.035+ |
| Qwen Image 2.0 Proserverless | $0.08 per image | Per image; docs print $0.075+ |
| Wan 2.6 Imageserverless | $0.03 per image | Per image; estimate, varies |
Hardware time
| Model | Price | Note |
|---|---|---|
| NVIDIA HGX H100dedicated inference | $5.49 per hour | Per GPU per hour, on-demand pay as you go; reserved = Contact sales |
| NVIDIA HGX B200dedicated inference | $8.99 per hour | Per GPU per hour, on-demand pay as you go |
| NVIDIA HGX H100GPU clusters | $3.99 per hour | Per GPU per hour, on-demand; preemptible $1.99; reserved $3.69 (3-30 d) / $3.45 (31-90 d) / $3.19 (91-180 d) |
| NVIDIA HGX H200GPU clusters | $5.99 per hour | Per GPU per hour, on-demand; preemptible $2.99; reserved $4.99 / $4.15 / $3.99 |
| NVIDIA HGX B200GPU clusters | $8.19 per hour | Per GPU per hour, on-demand; preemptible $4.09; reserved $7.99 / $7.79 / $6.79 |
| NVIDIA HGX B300GPU clusters | $9.99 per hour | Per GPU per hour, on-demand; preemptible $4.99; reserved = Contact us |
31 of these 51 prices are ranked in the index; the rest say why in their note. Text and embeddings per 1M tokens unless the row says otherwise.
Discounts and surcharges, as published
- Batch API: docs say 'up to 50% off serverless rates', but 'Selected serverless models run at 50% off batch rates' and the only model listed is openai/whisper-large-v3; 'Models not listed run at standard rates.' The pricing page has a 'Batch API price' toggle but shows no separate batch figures for chat models.
- Kimi K3's promotion ('Promotion valid until 10/11/2026' — $2.70 input / $13.50 output instead of $3.00 / $15.00) has ended: re-read 16 Oct, the page prints $3.00 / $0.30 cached / $15.00 with no PROMO tag.
- GPU Clusters: reserved per-GPU-hour rates for 3-30, 31-90, 91-180 days (e.g. HGX H100 $3.69 / $3.45 / $3.19) and preemptible compute (HGX H100 $1.99).
Where these numbers come from
Read on together.ai/pricing on 16 Oct 2026, checked against 3 more of Together AI’s own pages. Macrostack measures nothing and is paid by no provider on this index. Prices change; the date is the claim.
The other providers in the index
- The AI API price index — every provider, cheapest first, and the download
- OpenAI API pricing — model maker
- Anthropic API pricing — model maker
- Google Gemini API pricing — model maker
- Mistral AI API pricing — model maker
- Cohere API pricing — model maker
- Groq API pricing — open-model host
- Fireworks AI API pricing — open-model host
- Replicate API pricing — open-model host
- Deepgram API pricing — voice
- ElevenLabs API pricing — voice
Questions
- How much does the Together AI API cost?
- On 16 Oct 2026, Together AI's current text models cost $0.14 to $15.00 per million output tokens at the standard tier (Llama 3 8B Instruct Lite to Kimi K3). The full list, with input and cached-input prices, is on this page.
- Does Together AI have a free tier?
- What Together AI's own pages say: Pricing page only says 'Start for free, scale on demand' (page meta description); no credit amount. The official Credits doc says: 'Together AI does not currently offer free trials. Access to the Together platform requires a minimum $15 credit purchase.' Two chat models are priced at zero on the pricing page: Ternary Bonsai 27B ($0.00 / $0.00) and Tev1 4B Experimental (output 'Free').
- Does Together AI offer discounts?
- Batch API: docs say 'up to 50% off serverless rates', but 'Selected serverless models run at 50% off batch rates' and the only model listed is openai/whisper-large-v3; 'Models not listed run at standard rates.' The pricing page has a 'Batch API price' toggle but shows no separate batch figures for chat models. Kimi K3's promotion ('Promotion valid until 10/11/2026' — $2.70 input / $13.50 output instead of $3.00 / $15.00) has ended: re-read 16 Oct, the page prints $3.00 / $0.30 cached / $15.00 with no PROMO tag. GPU Clusters: reserved per-GPU-hour rates for 3-30, 31-90, 91-180 days (e.g. HGX H100 $3.69 / $3.45 / $3.19) and preemptible compute (HGX H100 $1.99).