OpenRouter
AI API Gateways
OpenRouter is the largest AI model gateway — 374+ models from Anthropic, OpenAI, Google, DeepSeek and more, with free tiers, BYOK support, and unified OpenAI-compatible API.
★★★★☆4.3
Docs-based · official check ← All tools
Best AI API Gateways in 2026: Pricing, Models & Speed
An AI API gateway is only as good as its model catalog, speed and free tier. OpenRouter is the breadth play: 374+ models from Anthropic, OpenAI, Google and DeepSeek behind one OpenAI-compatible key, with free variants of many models and BYOK support. Groq is the speed play — open-source LLMs on custom LPU chips returning 500+ tokens per second, with a free tier that needs no credit card. Together AI is the production play: 100+ open-source models, fine-tuning and dedicated GPU capacity, $5 free signup credit. DashScope is the Chinese-models play — Qwen, GLM, MiniMax and friends with native Chinese support and free token-credit promotions for new signups. Fireworks AI is the open-weight workhorse: 200+ models, self-serve post-training (SFT/LoRA/RL), 99.9% uptime SLA, pay-per-token with no GPU to provision, $1 free credit for new accounts (verified 2026-09-07 through 2026-09-11 against official pages).Fireworks AI is the open-weight workhorse in this lineup - 200+ models, self-serve post-training and a 99.9% uptime SLA with pay-per-token billing and no GPU to provision. It is the natural fit when you need reliable open-model inference plus fine-tuning without operating your own cluster. Pick OpenRouter for catalog breadth, Groq for raw speed and a no-card free tier, Together AI for production fine-tuning, DashScope for Chinese-language work, and Fireworks when open-weights reliability and self-serve training are the job. If your traffic is bursty, OpenRouter's free model tier keeps a prototyping path open, while Groq's batch and prompt-caching discounts cut paid inference roughly in half.
How we rated
- Model selection (breadth)
- Free tier generosity
- Inference speed
- Fine-tuning & production features
OpenRouter
Best overall & best model selection★★★★4.3
OpenRouter has by far the broadest model catalog — 374+ models including GPT, Claude, Gemini, DeepSeek, Llama and every major open model — all through one OpenAI-compatible API key. It supports BYOK (bring your own key) with 1 million free routing requests per month, and has free model variants with rate limits (20 req/min, 50-1000/day). The trade-off: quality varies across the marketplace, and some models are slower than direct API access.
Free tier: Yes
Paid from: $0.07/1M tokens (Pay-per-token, no monthly fees; 374+ models; free models: 20 req/min, 50 req/day (under $10 lifetime credits) or 1000/day ($10+); BYOK 1M free routing requests/mo; cheapest free models (DeepSeek R1, Llama 3.3 70B, Gemma 3, Gemini Flash) all rate-limited; paid models start at ~$0.07/1M input for small open models)
Groq
Fastest inference & best free tier for prototyping★★★★4.2
Groq runs open-source models on custom LPU chips that deliver 500+ tokens per second — the fastest inference API available. Its free tier is genuinely generous: 30 req/min, 6000 tokens/min, 14400 req/day, all models, no credit card required. Pricing starts at $0.05/1M input for Llama 3.1 8B, with batch and prompt caching discounts stacking to ~25% of on-demand. The catch: only open-source models — no GPT, Claude or Gemini.
Free tier: Yes
Paid from: $0.05/1M tokens (Pay-per-token, no monthly fees; free tier: 30 req/min, 6000 tokens/min, 14400 req/day, all models, no credit card; paid = Developer pay-as-you-go (rate limit increases with usage); cheapest: Llama 3.1 8B $0.05/$0.08 per 1M; flagship Llama 3.3 70B $0.59/$0.79 per 1M; Batch API + prompt caching each 50% off (~25% stacked))
Together AI
Best for production & fine-tuning★★★★3.9
Together AI is built for production AI teams — 100+ open models, LoRA and full fine-tuning, dedicated GPU capacity, and batch API with 50% off. Pricing starts at $0.05/1M tokens for small models. The trade-off: no permanent free tier (only $5 signup credit), no proprietary models, and the interface is more technical than OpenRouter. Best for teams that already know which open models they want to run in production.
Free tier: No — trial only · $5 free signup credit
Paid from: $0.05/1M tokens (Pay-per-token, no monthly minimums; $5 free signup credit (no credit card required for trial); cheapest: GPT-OSS 20B $0.05/$0.20 per 1M; Llama 3.3 70B $0.88 flat per 1M; DeepSeek V3.1 $0.60/$1.70; DeepSeek V4 Pro $2.10/$4.40; fine-tuning LoRA extra; Batch API 50% off; dedicated GPU hours separate)
DashScope
Best for Chinese models & Qwen-native work★★★★3.8
DashScope (Alibaba Cloud) specializes in Qwen, GLM, MiniMax and other Chinese models at very competitive prices — Qwen3.6-Flash from ~$0.03/1M input, with Qwen3.7-Plus and Qwen3.8-Max as higher tiers. It has full Chinese UI and documentation, and OpenAI-compatible endpoints. The catch: billing is in CNY through Alibaba Cloud, which adds friction for non-China teams, and the model catalog is China-focused rather than global.
Free tier: Yes · Free tier with token credits on new signup
Paid from: ¥0.20/1M tokens (~$0.03) (Pay-per-token in CNY; Qwen3.6-Flash ¥0.20/1M input / ¥1.20 output (~$0.03/$0.17); Qwen3.7-Plus ¥0.32/¥1.28 (~$0.05/$0.18); Qwen3.8-Max $2.00/$6.00 per 1M (GA Aug 2026); free signup credits vary by promotion (historical: free tier with RPM/daily caps); billing in CNY via Alibaba Cloud; full Chinese UI + docs)
★★★★4.0
Free tier: No — trial only · $1 free credit for new accounts
Paid from: $0.10/1M tokens (Pay-per-token for serverless; new accounts get only $1 in free credits (less than Together AI's $5 and Cerebras' 1M tokens/day); Serverless from $0.10/1M tokens for text & vision under 4B; On-Demand dedicated from $7.00/hr H100 (billed per second, no cold-boot cost); LoRA fine-tuning from $0.50/1M training tokens; Usage limit tier 1 = $50/month by default, hard-blocks requests when exceeded even with remaining credits)
Bottom line
Bottom line: OpenRouter is the best all-round pick if you want one API key for every model and you value breadth over everything else. Groq is the best pick for speed and prototyping — its free tier is genuinely useful and the LPU speed is unmatched. Together AI is the production platform choice for teams that need fine-tuning and dedicated capacity on open models. DashScope is the right choice if you primarily work with Qwen or Chinese-language models.
FAQs
Which AI API gateway has the most models?
OpenRouter has the broadest catalog with 374+ models from every major provider — Anthropic, OpenAI, Google, DeepSeek, Meta and many more. Groq and Together AI focus on open-source models only.
Which has the best free tier?
Groq has the most generous permanent free tier: 30 req/min, 14400 req/day, all models, no credit card required. OpenRouter also has free models but with tighter daily caps (50/day under $10 spend). DashScope signup credits vary by promotion.
Which is the fastest?
Groq is the fastest by a large margin — 500+ tokens per second on its custom LPU chips, which is 5-10x faster than GPU-based inference for comparable models.
Can I use fine-tuning?
Together AI has the strongest fine-tuning support with LoRA and full fine-tuning across many models. OpenRouter supports fine-tuning on some models. Groq and DashScope have more limited fine-tuning options.