Together AI
Together AI provides fast inference and fine-tuning infrastructure for open-source LLMs at scale.
Pros
- Full-stack AI platform with serverless, batch, provisioned, and dedicated inference options
- fine-tuning for open-source models
- GPU clusters from instant self-serve to thousands of GPUs
- claims 2x faster inference and 60% lower cost vs proprietary APIs
Cons
- No fixed pricing disclosed—token-based consumption model requires custom quote
- high barrier to entry without SaaS simplicity
- requires technical infrastructure knowledge
- startup program eligibility unclear
Cost
Serverless inference is billed per million tokens and varies by model: Ternary Bonsai 27B is free at $0.00 in and out, LFM2.5-8B-A1B is $0.03 input and $0.12 output, DeepSeek V4 Flash 0731 is $0.14 input and $0.28 output, Qwen3.8-2.4T-A95B is $2.00 input and $6.00 output, and Kimi K3 is $3.00 input and $15.00 output. On-demand GPU clusters run $3.99 per GPU-hour for HGX H100, $5.99 for H200 and $8.19 for B200, with reserved 181-day-plus rates of $3.99 for H200 and $6.79 for B200; dedicated inference is $5.49 per GPU-hour for H100, currently promoted at $3.99, and $8.99 for B200. Fine-tuning runs $0.48 to $8.00 per million tokens, the code interpreter is $0.03 per 60-minute session, and storage is $0.16 per GiB per month.
Verdict
Best for AI companies and researchers needing cost-effective inference at scale with fine-tuning and dedicated compute options.
Visit Together AI