Together AI

Together AI is a web and API platform for developers, ML engineers, startups, and enterprises that need GPU infrastructure for training, fine-tuning, and high-throughput inference on open-source models.

Key capabilities include:

• Serverless Inference API with OpenAI-compatible endpoints for integrating 200+ models across text, code, image, vision, audio, embeddings, and multimodal use cases (including LLaMA, DeepSeek, Qwen, Mistral, and FLUX).
• Performance-optimized inference via the Together Inference Engine and the ATLAS speculator system, plus techniques such as speculative decoding, quantization, and FP8 kernels to improve speed and unit economics.
• Fine-tuning workflows for full fine-tuning and LoRA, designed for task-specific models while keeping control of private training data.
• Deployment options spanning serverless, dedicated GPU endpoints (single-tenant available), and enterprise features such as HIPAA compliance.

Pricing: Usage-based. Inference is billed per million tokens (input/output). Fine-tuning is metered. Dedicated GPU endpoints are billed per minute; batch inference is offered at 50% lower cost for most models. Tiered plans include Scale (higher throughput, SLA) and Enterprise (private VPC, 99.9% SLA).

Notable for combining a large open-model catalog with production infrastructure and ongoing systems research, including access to frontier hardware (for example NVIDIA GB200 and GB300) and optimizations aimed at faster, lower-cost inference at scale.