Together AI is an inference platform that hosts hundreds of open models behind one OpenAI-compatible API. Its appeal was a single integration for four modalities instead of three or four separate signups. That is still true, but it is now a paid convenience rather than a free one. If you're already weighing Groq, Cerebras, Gemini, or DeepSeek, this is the one-key multimodal option.

What the Free Tier Used to Be

Until recently, models whose ID ended in -Free were callable without consuming credits, alongside a $1 signup credit for everything else. That is no longer the case: calls now fail until the account is funded, and no model carries a free suffix. The lineup below is kept because the model coverage is still accurate — only the price changed.

None of the -Free / -free model IDs above resolve any more. They are kept here as a reference for anyone migrating off an older tutorial: the suffix was removed along with the free tier. (Verified 2026-09-01.)

The $1 trial credit is enough to exercise dozens of paid models — Mixtral 8x22B, Qwen 2.5 72B, Llama 3.1 405B, Whisper, BGE/M2-BERT embeddings — for tens of thousands of tokens each before you commit a card.

Note: Together AI quietly retires and renames "Free" models over time. If a model ID stops working, check the official model list for the current Free variant.

Get a Key and Make Your First Call

Sign up at api.together.ai with email, Google, or GitHub. Signing up is still free and needs no card, but the account starts in read-only mode: inference calls fail until you make a deposit. From Settings → API Keys, copy your key and export it: export TOGETHER_API_KEY="your_key_here". The $1 credit and all -Free models activate immediately.

Together AI is fully OpenAI-compatible, so a one-shot curl confirms everything works:

curl https://api.together.xyz/v1/chat/completions 
  -H "Authorization: Bearer $TOGETHER_API_KEY" 
  -H "Content-Type: application/json" 
  -d '{
    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
    "messages": [
      {"role": "user", "content": "Explain pgvector in two sentences."}
    ]
  }'

In Python, either use the thin official SDK or point the OpenAI client at the Together base URL — a two-line swap from existing OpenAI code:

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["TOGETHER_API_KEY"],
    base_url="https://api.together.xyz/v1",
)

response = client.chat.completions.create(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free",
    messages=[{"role": "user", "content": "Write a haiku about caching."}],
)
print(response.choices[0].message.content)

Every OpenAI parameter — temperature, top_p, stop, response_format, tools, tool_choice — works identically. Streaming is real streaming on the Free tier too: set stream=True and tokens appear at roughly the model's true generation rate, usable for live chat UIs before you pay a cent.

Reasoning: DeepSeek R1 Distill

The 70B distilled R1 variant keeps most of the full model's reasoning at a fraction of the parameters, emitting a <think>...</think> block before its answer — show it for transparency or strip it for clean output.

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-R1-Distill-Llama-70B-free",
    messages=[{"role": "user", "content": "A bookstore sold 60 books Monday, then sales grew 12% each day through Friday. Total for the week? Show your work."}],
    max_tokens=2000,
)
print(response.choices[0].message.content)

Image Generation: FLUX.1 [schnell]

Black Forest Labs' fast text-to-image model, distilled to 4 sampling steps and Apache 2.0 licensed. For covers, social posts, and mockups it often beats SDXL on prompt adherence at a fraction of the inference time.

response = client.images.generate(
    # -Free suffix retired with the free tier; use
    # "black-forest-labs/FLUX.1-schnell" on a funded account
    model="black-forest-labs/FLUX.1-schnell-Free",
    prompt="Clean isometric illustration of an AI agent fetching data from a cloud database, soft pastels, no text",
    width=1024, height=1024, steps=4, n=1,
)
print(response.data[0].url)

Vision: Llama 3.2 Vision Free

Accepts standard OpenAI multimodal messages — text plus image URL or base64. The cheapest path in 2026 to a working "describe this screenshot" or "extract data from this chart" feature. Dense-document OCR still favors a paid vision model, but for screenshots, charts, and product photos it's genuinely useful.

response = client.chat.completions.create(
    model="meta-llama/Llama-Vision-Free",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What does this dashboard show? List the three highest values."},
            {"type": "image_url", "image_url": {"url": "https://example.com/dashboard.png"}}
        ]
    }],
)
print(response.choices[0].message.content)

Together AI vs Other Free APIs

Together AI wins on coverage — it's the only one here with chat, reasoning, vision, and image generation on one key. The others win on: raw speed (Cerebras, Groq), context window (Gemini's 1M tokens), or model variety (OpenRouter). Pick by what you're optimizing:

• Everything on one key? → Together AI Free
• Fastest response (sub-1s to first token)? → Cerebras (trial) or Groq
• 1M-token context? → Gemini
• Widest catalogue of free models? → OpenRouter
• Best free embedding + reranker for RAG? → Cohere
• Inference inside Cloudflare edge functions? → Cloudflare Workers AI

Rate Limits and Pricing

Free-tier limits keep costs predictable and are meant for development, not a production user base. Exact numbers live on the official rate limits page; as a 2026 mental model: -Free chat models get low double-digit requests/minute with smaller daily caps; -Free image models get tighter caps (often a few per minute); paid models on trial credit run at standard tier-1 limits, capped by your $1 budget. If a side project gets traction, move to paid or layer prompt caching in front.

When you outgrow free, serverless pricing is competitive and pay-as-you-go with no monthly minimum — and the same key covers both free and paid, so there's no migration when you flip one hot model to paid:

Check the official pricing page for current numbers.

Frequently Asked Questions

Is the Free tier really free, or a trial?

Neither, any more. There is no free tier: a new account is read-only until you make a deposit, and the -Free model IDs no longer exist. Together AI is now priced per token like any other host — competitively, but not at $0.

Do I need a credit card?

Yes. Without a funded balance the account stays in read-only mode and every inference call fails. If you need an OpenAI-compatible endpoint with no payment method at all, Groq still offers one (openai/gpt-oss-120b, 1,000 requests/day).

Is the API truly OpenAI-compatible?

Yes for chat completions, streaming, tool calling, and embeddings. Image generation uses Together's own endpoint shape (closely mirroring OpenAI's). In practice, point any OpenAI SDK at https://api.together.xyz/v1 and most code works unchanged.

How does it compare to running models locally with Ollama?

Ollama is unbeatable for offline, zero-cost tasks but bounded by your laptop's GPU — Llama 3.3 70B locally needs serious hardware. Together AI's Free tier runs the same model on a datacenter GPU, just rate-limited. They're complements: prototype locally on a small model, call Together for the 70B when it matters.

Final Verdict

Together AI's Free tier solves a problem most free APIs ignore: multimodal coverage on a single key. Where every rival is great at one thing and forces you to integrate three or four, Together gives you chat, reasoning, vision and image generation behind one endpoint, one SDK, one key. That convenience is real and the per-token pricing is competitive — but it is a paid platform now, so it no longer belongs in a $0 stack. If the budget is genuinely zero, cover chat with Groq and images with Cloudflare Workers AI, and accept two integrations instead of one.

Related Reads

• Free AI APIs in 2026: We Tested 10, Four Now Want a Card
• Groq vs Cerebras vs Gemini: Free AI API Speed Test
• DeepSeek API Is No Longer Free: We Checked the Balance
• OpenRouter Free Tier Tested: 18 Free Models, One API Key
• Cohere Free API: Embedding and Rerank for RAG