Together AI
MiniMax M3, Kimi K3, GLM-5.3, DeepSeek V4 Pro, gpt-oss-120B — plus GB200/B200 GPU clusters
Together AI offers hosted inference for open-weight models, GPU clusters, provisioned throughput and fine-tuning. Compare listed rates and check the provider for current models, availability and terms.
An open-weight model catalog — MiniMax M3, Kimi K3, GLM-5.3, DeepSeek V4 Pro, gpt-oss-120B — with GPU clusters (GB200/B200), Provisioned Throughput, and a Serverless Training API.
- Open-model catalog — MiniMax M3, Kimi K3, GLM-5.3, DeepSeek V4 Pro, gpt-oss-120B, Qwen3.x, Llama 4
- GB200/B200/H200/H100 GPU clusters with preemptible and reserved pricing
- Provisioned Throughput (PTUs) for SLA-backed capacity
- Serverless Training API for fine-tuning without infrastructure
- Fine-tuning for Llama 4, Qwen3.5, DeepSeek, GLM, and more
- OpenAI-compatible API — easy migration
- No proprietary frontier models
- Throughput lower than Groq/Cerebras for speed-critical apps
- Fine-tuning adds complexity vs. pure inference providers
Together AI Models
| Model | Modality | Tier | Price | Cached /1M | TTFT | Throughput | Context | Intel. | Caps | |
|---|---|---|---|---|---|---|---|---|---|---|
GLM-5.3-Flash GLM | Chat | budget | $0.15/ $0.50 in / out per 1M EST Oct 2026 | — | 200ms | 300 t/s | 131K | 68 | ||
gpt-oss-120B GPT-OSS | Chat | efficient | $0.15/ $0.60 in / out per 1M EST Oct 2026 | — | 280ms | 200 t/s | 131K | 74 | ||
Qwen3.8-Flash Qwen 3 | Chat | budget | $0.15/ $0.47 in / out per 1M EST Oct 2026 | — | 200ms | 280 t/s | 131K | 72 | ||
Qwen3.5-9B Qwen 3.5 | Chat | budget | $0.17/ $0.25 in / out per 1M EST Oct 2026 | — | — | — | 262K | 68 | ||
Llama-3.2-11B-Vision-Instruct-Turbo Llama 3.2 | Chat | budget | $0.18/ $0.18 in / out per 1M EST Jul 2026 | — | 180ms | 350 t/s | 131K | 48 | ||
Llama-4-Scout-17B-16E-Instruct Llama 4 | Chat | efficient | $0.18/ $0.59 in / out per 1M EST Jul 2026 | — | 200ms | 320 t/s | 131K | 68 | ||
Qwen3-235B-A22B-Instruct-2507-FP8-Throughput Qwen3 | Chat | advanced | $0.20/ $0.60 in / out per 1M EST Oct 2026 | — | — | — | 262K | 83 | ||
Llama-4-Maverick-17B-128E-Instruct-FP8 Llama 4 | Chat | advanced | $0.27/ $0.85 in / out per 1M EST Jul 2026 | — | 350ms | 200 t/s | 524K | 76 | ||
MiniMax-M3 MiniMax | Chat | efficient | $0.30/ $1.20 in / out per 1M EST Oct 2026 | — | 350ms | 180 t/s | 131K | 78 | ||
DeepSeek-V4.1-Flash DeepSeek V4 | Chat | efficient | $0.30/ $1.20 in / out per 1M EST Oct 2026 | — | 250ms | 220 t/s | 131K | 80 | ||
Qwen2.5-7B-Instruct-Turbo Qwen2.5 | Chat | budget | $0.30/ $0.30 in / out per 1M EST Oct 2026 | — | — | — | 131K | 55 | ||
Qwen3.7-Plus Qwen 3 | Chat | efficient | $0.32/ $1.28 in / out per 1M EST Oct 2026 | — | 300ms | 200 t/s | 131K | 82 | ||
gemma-4-31b Gemma 4 | Chat | efficient | $0.39/ $0.97 in / out per 1M EST Oct 2026 | — | 280ms | 220 t/s | 131K | 75 | ||
Qwen3.6-Plus Qwen 3 | Chat | advanced | $0.50/ $3.00 in / out per 1M EST Oct 2026 | — | — | — | 131K | 82 | ||
Qwen3-235B-A22B Qwen 3 | Chat | advanced | $0.60/ $0.60 in / out per 1M EST Jul 2026 | — | 450ms | 60 t/s | 41K | 82 | ||
Qwen3.5-397B-A17B Qwen 3.5 | Chat | advanced | $0.60/ $3.60 in / out per 1M EST Oct 2026 | $0.35 | — | — | 262K | 88 | ||
Llama-3.3-70B-Instruct-Turbo Llama 3.3 | Chat | efficient | $1.04/ $1.04 in / out per 1M EST Jul 2026 | — | 290ms | 100 t/s | 131K | 74 | ||
Mixtral-8x22B-Instruct-v0.1 Mixtral | Chat | advanced | $1.20/ $1.20 in / out per 1M EST Jul 2026 | — | 400ms | 70 t/s | 66K | 74 | ||
DeepSeek-V3 DeepSeek V3 | Chat | advanced | $1.25/ $1.25 in / out per 1M EST Jul 2026 | — | 380ms | 75 t/s | 131K | 80 | ||
DeepSeek-V4-Pro-0813 DeepSeek V4 | Chat | advanced | $1.32/ $3.96 in / out per 1M EST Oct 2026 | — | 450ms | 130 t/s | 131K | 88 | ||
GLM-5.3 GLM | Chat | advanced | $1.40/ $4.40 in / out per 1M EST Oct 2026 | — | 420ms | 140 t/s | 131K | 82 | ||
Qwen3.7-Max Qwen 3 | Chat | advanced | $1.50/ $4.50 in / out per 1M EST Oct 2026 | $0.30 | — | — | 131K | 84 | ||
Qwen3.8-2.4T-A95B Qwen 3 | Chat | frontier | $2.00/ $6.00 in / out per 1M EST Oct 2026 | — | 500ms | 80 t/s | 131K | 90 | ||
Kimi-K3 Kimi | Chat | frontier | $3.00/ $15.00 in / out per 1M EST Oct 2026 | — | 600ms | 90 t/s | 131K | 88 | ||
Meta-Llama-3.1-405B-Instruct-Turbo Llama 3.1 | Chat | advanced | $3.50/ $3.50 in / out per 1M EST Jul 2026 | — | 900ms | 60 t/s | 131K | 72 |
Together AI — Frequently Asked Questions (FAQs)
Similar Providers
Together AI Reviews
All LLM Providers
Full comparison tableTogether AI LLM pricing overview
Together AI is an inference API provider that hosts open-weight and third-party large language models. Together AI offers hosted inference for open-weight models, GPU clusters, provisioned throughput and fine-tuning. Compare listed rates and check the provider for current models, availability and terms. Common use cases include Open-source, Fine-tuning, Coding, Chat. Headquartered in San Francisco, CA, founded 2022. All models are accessible via a REST API compatible with standard OpenAI-style request formats, enabling drop-in integration with most LLM frameworks and orchestration tools.
Together AI vs other LLM providers
Together AI competes with OpenAI, Anthropic, Google, Groq, Together AI, Mistral, Cohere, and other inference API providers across dimensions of price, throughput, latency, context window, and model intelligence. An open-weight model catalog — MiniMax M3, Kimi K3, GLM-5.3, DeepSeek V4 Pro, gpt-oss-120B — with GPU clusters (GB200/B200), Provisioned Throughput, and a Serverless Training API. Use the LLM provider comparison tool to see Together AI token pricing, latency, and throughput side-by-side with any other provider. The full LLM pricing table shows all providers ranked by input token cost, output token cost, and throughput in a single sortable view.
Why choose Together AI?
Together AI stands out because: An open-weight model catalog — MiniMax M3, Kimi K3, GLM-5.3, DeepSeek V4 Pro, gpt-oss-120B — with GPU clusters (GB200/B200), Provisioned Throughput, and a Serverless Training API.. The API currently offers 25 models, spanning 2 frontier, 11 advanced, 7 efficient, 5 budget tiers. For teams running high-volume inference workloads, prompt caching and batch API endpoints can reduce effective input token costs by 50–90% — check the context window cost guide for a full breakdown of caching economics.
Understanding Together AI token pricing
Together AI charges separately for input (prompt) and output (completion) tokens, priced per 1M tokens in USD. Output tokens are typically 3–5× more expensive than input tokens due to the compute cost of autoregressive generation. The cheapest {meta.name} input rate is currently $0.15/1M tokens. Prices shown are sourced from Together AI's public pricing page and updated every 15 minutes. Need help estimating your spend? Read the LLM API cost calculator guide — it covers tokens, context windows, prompt caching, and batch discounts with worked examples for RAG, chat history, and document processing workloads.
Together AI context window and model capabilities
Context window size directly affects both capability and cost — every token in the context window is charged as an input token. Together AI's largest context window is 524K tokens. For RAG and document processing workloads, longer context windows enable richer retrieval but increase per-call costs proportionally. Prompt caching — where supported — stores the KV state of repeated prefixes and charges 75–90% less for cache hits, making it the most impactful cost optimization for applications with consistent system prompts or retrieved documents. See the LLM context window cost guide for a full analysis of how context length affects your API bill.
Self-hosted vs managed inference: when Together AI makes sense
Managed inference APIs like Together AI eliminate infrastructure overhead — no GPU provisioning, driver management, or model serving stack to maintain. The trade-off is cost at scale: a single H100 at ~$2.50/hr can serve ~500K tokens/min of Llama 3.3 70B, which at Together AI API rates would cost significantly more per token. The break-even point depends on your request volume, latency requirements, and engineering capacity. For teams processing fewer than ~10M tokens/day, managed APIs are almost always cheaper when total cost of ownership is considered. Above that threshold, self-hosted inference on rented GPU compute typically wins on unit economics. Read the cheapest GPU cloud guide for a full break-even analysis. Historical Together AI token price data is available in the LLM price history charts.
Compare Together AI With Another LLM Provider
Compare Together AI vs other providers
Side-by-side token pricing, latency, throughput, and context window. View all LLM provider comparisons →
Looking for alternatives?
Compare Together AI with other LLM API providers side by side.