AI/ML API (AIMLAPI)
Listed at https://aimlapi.com
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | Per-1M-token list rates, many below first-party via aggregator volume discounts (verified aimlapi.com model pages 2026-09-07): Claude Sonnet 5 $2.60, GPT-5.2 $2.275, DeepSeek V4 Pro $0.5655, Gemini 3.6 Flash $1.95, Claude Fable 5 $13 | per million tokens |
| Output | Claude Sonnet 5 $13.00; GPT-5.2 $18.20; DeepSeek V4 Pro $1.131; Gemini 3.6 Flash $9.75; Claude Fable 5 $65 | per million tokens |
🤖 Supported Models (785)
✨ Pros
- ✓One OpenAI-compatible key covers 785+ unique models / 1000+ marketed: Claude Sonnet 5, GPT-5.2, Gemini 3.6 Flash, Grok, DeepSeek V4 Pro, Qwen, GLM, Kimi + image/video/TTS/STT/embedding/OCR/batch
- ✓Broad discounting: flagships list at/below first-party (DeepSeek V4 Pro $0.5655/$1.131, Claude Sonnet 5 $2.60/$13.00)
- ✓Batch ~50% off realtime for offline bulk inference
- ✓Live & active: /v1/models HTTP 200 (937 rows / 785 unique 2026-09-07); HN presence since 2024-09; marketing current to 2026-09
- ✓EU angle: Estonia AIMLAPI OÜ entity + GDPR positioning friendly to European developers
⚠️ Cons
- ×Free tier long paused: no ongoing allowance, only signup trial credits
- ×Opaque region: single 'global endpoint' with no published DC/region map, no latency or residency SLA
- ×Discounts not guaranteed: prices anchor to first-party list, can rise on upstream re-tiering
- ×Some cache/batch figures inconsistent across docs (JSON-LD single-price vs page) — trust the model page
- ×Mainland China needs proxy; no CN payment or residency
🎯 Best For
Developers, agent teams and EU/go-global teams wanting one OpenAI-compatible key across the Claude/GPT/Gemini/Grok/DeepSeek/Qwen/GLM/Kimi family with discount pricing on flagships; anyone doing batch-discount offline inference or multi-vendor A/B without opening five accounts. Not for self-hosted-first shops needing fixed owned region + hard residency SLA, or direct mainland-China production.
💰 Pricing & Plans
| Model | Input ($/M) | Output ($/M) | Cached in ($/M) | Notes |
|---|---|---|---|---|
| Claude Sonnet 5 | $2.60 | $13.00 | $0.26 | 1M ctx; below Anthropic first-party (~$3/$15) |
| GPT-5.2 | $2.275 | $18.20 | $0.2275 | 400K ctx; reasoning/agent flagship |
| DeepSeek V4 Pro | $0.5655 | $1.131 | $0.0047 | 1M ctx; very low cached read |
| Gemini 3.6 Flash | $1.95 | $9.75 | $0.195 | 1.05M ctx; low-cost Flash tier |
| Claude Fable 5 | $13.00 | $65.00 | — | top frontier; list at first-party |
| Qwen / GLM / Kimi open | varies | varies | — | open-weights incl. Qwen3.8-Max class |
| Batch (async) | ~50% of realtime | — | — | offline bulk inference window pricing |
🔧 API & Developer Experience
- •OpenAI-Compatible Base URL: Drop-in endpoint at https://api.aimlapi.com/v1 (chat via /v1/chat/completions). Existing OpenAI SDKs/curl work by swapping base_url + api key from the account ('aimlapi.com/app/keys'). Docs show OpenAI(base_url='https://api.aimlapi.com/v1', api_key=...).
- •Named-Space Model IDs: Model identifiers are namespaced (anthropic/claude-sonnet-5, openai/gpt-5.2, deepseek/deepseek-v4-pro) so vendors stay unambiguous behind one endpoint.
- •Docs + GitHub: GitBook docs at docs.aimlapi.com with Markdown export (append .md) + llms.txt + an agent-friendly 'ask' query endpoint; GitHub org + Discord active.
- •No Free Tier / Top-up: Official FAQ: 'The Free Tier is currently paused.' Signup trial credits only; then credit-based prepaid billing per 1M tokens.
- •Batch + Gen-Media: Async batch endpoints (~50% below realtime) for bulk; plus text-to-image/video, TTS/STT, embeddings, OCR on the same key.
- •OpenWeights + Proprietary: Unlike pure open-weight hosts, AIMLAPI also resells Claude/GPT/Gemini/Grok — a practical alternative menu to OpenRouter-style aggregators.
⚡ Negotiated Discounts on One Key
What makes AIMLAPI interesting is its economic model: it is a hosted resale aggregator that negotiates volume rates on proprietary flagships and re-prices them below first-party list. The live catalog exposes 785 unique model IDs under one endpoint, with namespaced identifiers (anthropic/claude-sonnet-5, openai/gpt-5.2, deepseek/deepseek-v4-pro) so vendor ambiguity disappears. Instead of maintaining five accounts and five keys across OpenAI, Anthropic, Google, xAI and DeepSeek, you route everything through api.aimlapi.com/v1 and clear one prepaid balance. Batch endpoints run at roughly half the realtime rate for offline workloads, and Gen-Media (image, video, TTS, OCR) rides the same key. That one-key-everything surface plus built-in discounting is the whole product — closer to OpenRouter with aggressive negotiated markdowns than to a self-hosted gateway like LiteLLM.
🌐 Regional Availability & Latency
AIMLAPI serves everything through one global OpenAI-compatible endpoint (https://api.aimlapi.com/v1) fronted by Cloudflare, and it does not publish a data-center or region map, so latency and data residency have no per-region SLA. The legal entities are AIMLAPI OÜ (Estonia) and Boiler Labs FZ-LLC (Dubai), which gives EU-based developers a reasonable GDPR reading — traffic terminates on Cloudflare's edge and is processed by an EU-incorporated operator. There is no mainland-China endpoint, no China payment method and no China data-residency option, so mainland production traffic must route through a proxy or a relay. For user-facing mainland-China traffic that is latency- or residency-sensitive, pair a domestic provider (Qwen, Zhipu, Doubao) as the China-facing layer, as with any overseas aggregator.