Together AI API Pricing and Model Catalog: Use-Case Guide
Together AI offers a broad catalog of open-weight models with fine-tuning support. Here's how to use their API cost-effectively.
Pricing data sourced from our catalog. Check data sources for provenance and freshness.
What is Together AI?
Together AI is an inference and fine-tuning platform specializing in:
- Open-weight models: Llama, Qwen, DeepSeek, Mistral, and more
- Fine-tuning: Train custom models on your data
- Inference: Optimized serving for open-weight models
- Developer-friendly: OpenAI-compatible API
Together AI model catalog
Chat models
| Model | Input | Output | Context |
|---|---|---|---|
| together-ai-up-to-4b | $0.1000 | $0.1000 | - |
| gpt-oss-20b | $0.0500 | $0.2000 | - |
| Meta-Llama-3.1-8B-Instruct-Turbo | $0.1800 | $0.1800 | - |
| together-ai-4.1b-8b | $0.2000 | $0.2000 | - |
| together-ai-8.1b-21b | $0.3000 | $0.3000 | 1K |
| gpt-oss-120b | $0.1500 | $0.6000 | 131K |
| Llama-4-Scout-17B-16E-Instruct | $0.1800 | $0.5900 | - |
| Qwen3-235B-A22B-fp8-tput | $0.2000 | $0.6000 | - |
Embedding models
| Model | Input |
|---|---|
| together-ai-embedding-up-to-150m | $0.0080 |
| bge-base-en-v1.5 | $0.0080 |
| bge-base-en-v1.5 | $0.0080 |
| together-ai-embedding-151m-to-350m | $0.0160 |
Cost estimation example
Let's estimate costs for a typical Together AI chat workload:
Model selection guidance
- Budget tasks: Choose smaller models like Llama 3.1 8B or GPT-OSS-20B for simple classification, extraction, or summarization
- General-purpose: Llama 3.3 70B, Qwen3-235B-A22B, or DeepSeek-V3.1 offer strong performance at moderate cost
- Reasoning-heavy: DeepSeek-R1, Qwen3-235B-A22B-Thinking, or Kimi-K2.5 for multi-step reasoning and analysis
- Code generation: Qwen3-Coder-480B-A35B-Instruct-FP8 or DeepSeek-V3.1 for coding workloads
- Embedding: BGE-base-en-v1.5 for vector search and RAG applications
Together AI strengths
- Open-weight models: Wide selection of community and proprietary open-weight models
- Fine-tuning: Train custom models on your data via the API
- Cost-effective inference: Competitive pricing for large open-weight models
- OpenAI-compatible API: Easy to migrate from OpenAI
- No vendor lock-in: Same models available elsewhere
Cost optimization tips
- Use smaller models for simple tasks: Route easy requests to cheaper models
- Leverage fine-tuning: A smaller fine-tuned model can outperform a larger general model
- Batch processing: Process multiple requests together for throughput
- Monitor token usage: Track input/output ratios to optimize prompts
- Use the cheapest model that works: Start with budget models and upgrade only when quality requires it
Compare Together AI models
Ready to compare Together AI models side by side? Use our tools:
Related guides
Open-weight inference provider pricing comparison
Compare Groq, Fireworks, DeepInfra, and Together AI for open-weight inference.
Cross-provider pricing comparison
How pricing compares across OpenAI, Anthropic, Google, Mistral, and DeepSeek.
Building a RAG application
How to choose models and optimize costs for RAG applications.
Hidden costs of LLM APIs
Rate limits, latency, evaluation overhead, and vendor risk beyond per-token pricing.
Frequently asked questions
What is Together AI's cheapest model?
Together AI's GPT-OSS-20B is typically the most cost-effective chat model. For embedding, BGE-base-en-v1.5 is the cheapest option.
Does Together AI support fine-tuning?
Yes. Together AI offers fine-tuning for many open-weight models, allowing you to train custom models on your data via the API.
How does Together AI compare to Fireworks AI?
Both offer open-weight inference. Together AI has a broader model catalog and fine-tuning support, while Fireworks AI focuses on low-latency inference.
Is the Together AI API compatible with OpenAI?
Yes. Together AI provides an OpenAI-compatible API, so you can swap endpoints with minimal code changes.