Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Fine-tuning - verified

Together AI fine-tuning: status, methods and prices

Offered. 3 offerings recorded: LoRA, DPO and Full. Training is priced per 1M training tokens, in Together AI's own units. Weight export: Yes.

Recorded as: Together AI.

Read from docs.together.ai, together.ai. Compare every fine-tuning API

Status
Offered
Offerings
3
Methods
LoRA, DPO, Full
Export weights
Yes

Status

Is Together AI fine-tuning still available?

Offered

Offered

All 3 offerings

Prices are in the provider's own unit, as written. Not stated means the page we checked does not say.
LoRANot stated

LoRA supervised fine-tuning

LoRA (adapter) fine-tuning

InferredThe pricing page shows two unlabeled tables under LoRA and Full fine-tuning tabs. We read the first as LoRA and the second as full fine-tuning by tab order, because the second table's models match the docs' full fine-tuning list.

Status
Not stated
Tunable base models
  • deepseek-ai/DeepSeek-V4-Flash-0731
  • deepseek-ai/DeepSeek-V3.1
  • zai-org/GLM-5.1
  • Qwen/Qwen3.8-27B
  • Qwen/Qwen3.5-397B-A17B
  • moonshotai/Kimi-K2.6
  • google/gemma-4-31B-it
  • openai/gpt-oss-20b
  • openai/gpt-oss-120b
  • meta-llama/Llama-4-Maverick-17B-128E-Instruct
  • meta-llama/Llama-3.3-70B-Instruct-Reference
  • meta-llama/Meta-Llama-3.1-8B-Instruct-Reference
  • mistralai/Mixtral-8x7B-Instruct-v0.1

Subset of the 29-row LoRA table; four quoted.

“deepseek-ai/DeepSeek-V3.1”
docs.together.ai
“openai/gpt-oss-120b”
docs.together.ai
“meta-llama/Meta-Llama-3.1-8B-Instruct-Reference”
docs.together.ai
“moonshotai/Kimi-K2.6”
docs.together.ai
Training price
LoRA supervised fine-tuning training prices in Together AI's own units
ModelPriceUnitQuote
Llama 3.1 8B$0.34 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.1 8B · $0.34 · $0.84 · $4.00”together.ai
Llama 3.3 70B$2.03 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.3 70B · $2.03 · $5.08 · $4.00”together.ai
gpt-oss-20B$0.40 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“gpt-oss-20B · $0.40 · $1.00 · $4.00”together.ai
gpt-oss-120B$2.50 (min charge $6.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“gpt-oss-120B · $2.50 · $6.25 · $6.00”together.ai
DeepSeek-V3.1$7.00 (min charge $20.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“DeepSeek-V3.1 · $7.00 · $17.50 · $20.00”together.ai
Kimi K2.6$15.00 (min charge $60.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Kimi K2.6 · $15.00 · $37.50 · $60.00”together.ai
GLM-5.1$40.00 (min charge $60.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“GLM-5.1 · $40.00 · $100.00 · $60.00”together.ai
Tuned-model inference
Not stated
Hosting requirement

dedicated endpoint (some tunable models cannot be deployed)

“Some models can be fine-tuned but cannot be deployed as dedicated endpoints.”
docs.together.ai
Export weights

Yes

“Serve your fine-tuned model on a dedicated endpoint or download it for local use.”
docs.together.ai
Data limits
Not stated
DPONot stated

Preference fine-tuning (DPO), LoRA

Direct preference optimization (preference tuning)

InferredThe pricing page shows two unlabeled tables under LoRA and Full fine-tuning tabs. We read the first as LoRA and the second as full fine-tuning by tab order, because the second table's models match the docs' full fine-tuning list.

Status
Not stated
Tunable base models
  • same as LoRA table (DPO context lengths listed per model)
“Context lengths are the maximum for that model in SFT and DPO modes.”
docs.together.ai
Training price
Preference fine-tuning (DPO), LoRA training prices in Together AI's own units
ModelPriceUnitQuote
Llama 3.1 8B$0.84per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.1 8B · $0.34 · $0.84 · $4.00”together.ai
Llama 3.3 70B$5.08per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.3 70B · $2.03 · $5.08 · $4.00”together.ai
gpt-oss-120B$6.25per 1M training tokens (dataset tokens x epochs + eval tokens)“gpt-oss-120B · $2.50 · $6.25 · $6.00”together.ai
Llama 3.3 70B (full fine-tuning DPO)$5.60per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.3 70B · $2.24 · $5.60 · $4.00”together.ai
Tuned-model inference
Not stated
Hosting requirement

dedicated endpoint (some tunable models cannot be deployed)

“Some models can be fine-tuned but cannot be deployed as dedicated endpoints.”
docs.together.ai
Export weights

Yes

“Serve your fine-tuned model on a dedicated endpoint or download it for local use.”
docs.together.ai
Data limits
Not stated
FullNot stated

Full-parameter supervised fine-tuning

Full-parameter fine-tuning

InferredThe pricing page shows two unlabeled tables under LoRA and Full fine-tuning tabs. We read the first as LoRA and the second as full fine-tuning by tab order, because the second table's models match the docs' full fine-tuning list.

Status
Not stated
Tunable base models
  • Qwen/Qwen3.8-27B
  • Qwen/Qwen3.5-9B
  • Qwen/Qwen3.5-4B
  • Qwen/Qwen3.6-27B
  • Qwen/Qwen3.5-27B
  • google/gemma-4-31B-it
  • google/gemma-4-31B-it-VLM
  • meta-llama/Llama-3.3-70B-Instruct-Reference
  • meta-llama/Meta-Llama-3.1-8B-Instruct-Reference
  • mistralai/Mixtral-8x7B-Instruct-v0.1
“Meta · Llama 3.3 70B Instruct Reference · meta-llama/Llama-3.3-70B-Instruct-Reference · 24576 · 12288 · 32 · 32 · 32”
docs.together.ai
“Qwen · Qwen3.8 27B · Qwen/Qwen3.8-27B · 131072 · 65536 · 4 · 4 · 4”
docs.together.ai
Training price
Full-parameter supervised fine-tuning training prices in Together AI's own units
ModelPriceUnitQuote
Llama 3.1 8B$0.38 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.1 8B · $0.38 · $0.94 · $4.00”together.ai
Llama 3.3 70B$2.24 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Llama 3.3 70B · $2.24 · $5.60 · $4.00”together.ai
Qwen3.5 27B$1.16 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Qwen3.5 27B · $1.16 · $2.88 · $4.00”together.ai
Gemma 4 31B$1.16 (min charge $4.00)per 1M training tokens (dataset tokens x epochs + eval tokens)“Gemma 4 31B · $1.16 · $2.88 · $4.00”together.ai
Tuned-model inference
Not stated
Hosting requirement

dedicated endpoint (some tunable models cannot be deployed)

“Some models can be fine-tuned but cannot be deployed as dedicated endpoints.”
docs.together.ai
Export weights

Yes

“Serve your fine-tuned model on a dedicated endpoint or download it for local use.”
docs.together.ai
Data limits
Not stated

Research notes

Pricing page has two unlabeled tables under 'LoRA fine-tuning' and 'Full fine-tuning' tabs; mapping first table = LoRA and second = Full is inferred from tab order and from the second table's model set matching the docs' full fine-tuning list. Each job has a per-model minimum charge.

Plain language

The methods, explained

SFTSupervised fine-tuning
You show the model example prompts with the answers you want, and it learns to answer like that. The default method almost everywhere.
DPODirect preference optimization (preference tuning)
You give pairs of answers to the same prompt, one preferred and one not, and the model learns to lean toward the preferred kind. Good for tone and style.
RFTReinforcement fine-tuning
The model attempts tasks and a grader (your code, or another model) scores each attempt; training pushes it toward higher-scoring answers. Often billed by training hour, plus the grader's tokens.
Vision SFTVision fine-tuning
Supervised fine-tuning where the examples include images, so the model gets better at a visual task.
DistillationDistillation
A large teacher model writes answers that are used to fine-tune a smaller, cheaper student model. You pay for the teacher's calls as well as the training.
LoRALoRA (adapter) fine-tuning
Trains a small set of extra weights, an adapter, on top of the frozen base model. Cheaper and faster than full fine-tuning; the adapter is served alongside the base model.
FullFull-parameter fine-tuning
Updates every weight in the model. Costs more than LoRA and produces a complete copy of the model.
CPTContinued pre-training
Keeps training the model on a large body of raw, unlabeled text from your domain, so it absorbs vocabulary and facts before any instruction tuning.

FAQ

Does Together AI offer fine-tuning?

Offered. 3 offerings: LoRA, DPO and Full.

How much does Together AI fine-tuning cost?

Per 1M training tokens: Llama 3.1 8B $0.34 (min charge $4.00), Llama 3.3 70B $2.03 (min charge $4.00), gpt-oss-20B $0.40 (min charge $4.00), gpt-oss-120B $2.50 (min charge $6.00) and 11 more. Prices are in Together AI's own units and are not converted.

Can you download a model you fine-tune on Together AI?

Yes. The docs say: "Serve your fine-tuned model on a dedicated endpoint or download it for local use."

Sources

Last verified . This is a directory of what each provider's own pages say, not a quality ranking. Prices are copied in the provider's own unit and never converted, so rows in different units are not comparable.

Newsletter

Know when a provider changes its fine-tuning terms

One email per week. New model releases, deprecations and pricing changes, and what they mean for what you build.