Together
Generative AI infrastructure software
Generative AI software
Large language model operationalization (LLMOps) software
- Features
- Ease of use
- Ease of management
- Quality of support
- Affordability
- Market presence
Take the quiz to check if Together and its alternatives fit your requirements.
$5 minimum credit purchase
Free Trial unavailable
Free version
Small
Medium
Large
-
What is Together
Together (often referred to as Together AI) is a platform for running and operationalizing large language models, combining model hosting/inference, fine-tuning, and APIs for building generative AI applications. It targets engineering and ML teams that need to deploy open-source or custom models for chat, RAG, and other text-generation workloads. The product emphasizes managed inference endpoints and training/fine-tuning workflows that can be integrated into application stacks. It is typically used as an alternative to building and operating GPU infrastructure and model-serving pipelines in-house.
Managed model inference APIs
Together provides hosted inference endpoints that teams can call from applications without standing up their own model-serving stack. This reduces the operational work associated with provisioning GPUs, deploying model servers, and handling scaling for common LLM workloads. It fits teams that want a production API surface for open-source and custom models. It also supports common integration patterns used in LLM application development.
Fine-tuning and training support
The platform supports fine-tuning workflows so teams can adapt base models to domain-specific data and tasks. This can shorten the path from experimentation to a deployable, versioned model endpoint. For organizations that lack internal training infrastructure, it provides a managed option for running these jobs. It aligns with LLMOps needs such as repeatable training runs and deployment of tuned artifacts.
Focus on open models
Together is oriented around using and operationalizing open-source model families rather than only proprietary models. This can help teams that need more control over model choice, weights, and deployment options. It also supports use cases where organizations want to avoid lock-in to a single closed model provider. The approach is relevant for teams building internal assistants, RAG systems, and custom text-generation services.
Limited end-to-end app tooling
Compared with platforms that bundle broader data preparation, analytics, and application orchestration, Together is more centered on model training and serving. Teams may still need separate tools for dataset governance, feature pipelines, evaluation harnesses, and application-layer observability. This can increase integration work for organizations seeking a single consolidated environment. The product is typically one component in a larger AI stack.
Model quality depends on choices
Outcomes depend heavily on the selected base model, fine-tuning data quality, and prompt/RAG design. Organizations may need in-house expertise to choose models, set safety/quality constraints, and run evaluations across versions. This is a common challenge when operationalizing open models at scale. The platform does not eliminate the need for rigorous testing and monitoring.
Cost and capacity variability
GPU-backed inference and training costs can vary significantly with model size, throughput requirements, and concurrency. Teams with spiky traffic or strict latency SLAs may need careful capacity planning and performance testing. Budget predictability can be harder than with smaller, fixed-scope AI features embedded in other business software. Procurement may require deeper technical validation to estimate ongoing spend.
Plan & Pricing
Pricing model: Pay-as-you-go (usage-based)
Free tier / trial: No platform-wide free trial. (See notes below about a product-specific free "Build (free)" Code Sandbox plan.)
Serverless inference (text & vision)
- Price units: per 1M tokens (input / output shown where provided).
- Example model prices (per 1M tokens): Llama 4 Maverick — $0.27 (input) / $0.85 (output); Qwen3.5-397B — $0.60 / $3.60; gpt-oss-120B — $0.15 / $0.60; Llama 3.2 3B Instruct Turbo — $0.06 / $0.06.
Image models
- Price units: per megapixel (MP).
- Examples: FLUX.1 Krea (dev) — $0.025/MP; Google Imagen 4.0 Preview — $0.04/MP; SD XL — $0.0019/MP.
Audio / Transcription / Embeddings / Moderation / Rerank
- Audio synthesis: per 1M characters (example: Cartesia Sonic-2 — $65/1M chars).
- Transcription: per audio minute (example: Whisper Large v3 — $0.0015 per minute).
- Embeddings: per 1M tokens (example: BGE-Base-EN v1.5 — $0.01/1M tokens; BGE-Large-EN v1.5 — $0.02).
- Moderation / rerank: per 1M tokens (examples shown on pricing page).
Batch API / Image / Video / Other
- Batch API: model-specific per‑1M token pricing (see Serverless inference table).
- Video generation: price per video (examples: Google Veo 3.0 + Audio — $3.20; MiniMax 01 Director — $0.28).
Fine-tuning
- Price units: per 1M tokens processed (training dataset size * epochs + eval tokens).
- Standard fine-tuning examples (per 1M tokens): Up to 16B — LoRA $0.48; Full FT $0.54. 17B–69B — LoRA $1.50; Full FT $1.65. 70–100B — LoRA $2.90; Full FT $3.20.
- Specialized model minimum charges and higher per-1M costs are listed for certain models (e.g., DeepSeek, GLM, Kimi, Llama 4 variants).
Dedicated Endpoints & GPU Cloud
- Dedicated endpoints (single-tenant GPU instances) price/hour examples: 1x H200 141GB — $4.99/hr; 1x H100 80GB — $3.36/hr; 1x A100 SXM 80GB — $2.56/hr.
- Instant Clusters (hourly per GPU): NVIDIA HGX H100 SXM — $2.99/hr (hourly rate shown; discounted rates for longer reservations available); H200 — $3.79/hr; B200 — $5.50/hr.
- Reserved clusters / Frontier AI Factory: contact sales / custom pricing for large-scale deployments.
Code execution / Code Sandbox
- VM credit pricing: $0.01486 per VM credit (one credit = base unit; VM sizes consume credits/hour).
- VM sizes (credits/hour and $/hr examples): Pico — 5 credits ($0.0743/hr); Nano — 10 credits ($0.1486/hr); Micro — 20 credits ($0.2972/hr); XLarge — 320 credits ($4.7552/hr).
- Concurrent VMs / plans: Build (free) plan — 10 concurrent VMs; Scale plan — 250 concurrent VMs (Scale plan base price and included free VM credits referenced in docs); Enterprise — custom.
Storage
- Shared filesystem: $0.16 per GiB per month.
Other notes
- Displayed prices refer to default resolution/duration; actual costs may vary by model settings.
- Many prices are listed per 1M tokens, per MP, per minute, or per video as appropriate. See vendor pricing page for model-by-model detail.
