AI Providers Knowledgebase
Catalog of first-party labs, cloud platforms, inference hosts, specialized APIs, and routers/gateways. Snapshot dated September 2026.
This skill is for selection and comparison. For SDK code, use the vendor skill:
Freshness rule
Prices, model IDs, and retention policies change without notice.
- Treat numbers in this skill as a dated snapshot (Sep 2026).
- Before quoting a production price or pinning a model ID, re-fetch the official pricing/docs URL in the provider entry.
- Prefer the vendor’s own docs over OpenRouter for first-party APIs. Prefer OpenRouter’s live listing for which hosts currently serve a model.
- Do not invent model IDs from training memory. If this catalog and a live page disagree, use the live page.
Live marketplace: OpenRouter Providers · OpenRouter Models
When to use which path
Need first-party features (Responses, Computer Use, native tools, enterprise DPA)
→ Call the lab or its cloud SKU (Azure / Vertex / Bedrock).
Need lowest latency on open weights
→ Groq (LPU) or Cerebras (wafer-scale); SambaNova for high TPS.
Need cheapest tokens on a popular open model
→ Compare DeepInfra, Novita, SiliconFlow, Parasail on OpenRouter for that model ID.
Need one key, many labs, automatic failover
→ Hosted router: OpenRouter (global), EUrouter (EU residency), Vercel AI Gateway (Vercel apps).
Need VPC, virtual keys, budgets, guardrails on *your* provider keys
→ Self-hosted / enterprise gateway: LiteLLM, Portkey, Kong, TrueFoundry.
Need a chat / coding / tool-calling agent
→ chat-agent-models.md (GPT-6, Claude 5, Gemini 3.x, Grok, DeepSeek, Kimi, GLM, …).
Need image generation or editing
→ image-models.md (GPT Image, FLUX.2, Nano Banana, Recraft, Seedream, Grok Imagine, Krea).
Need video generation
→ video-models.md (Sora 2 Pro, Veo 3.1, Runway, Kling, Hailuo, Wan, Seedance, HeyGen).
Need music or speech
→ audio-music-models.md (Lyria 3 music; Fish/MiniMax/Qwen/Deepgram TTS; gpt-audio; ASR).
Privacy cheat sheet (OpenRouter-reported, Sep 2026)
OpenRouter’s provider table is the fastest policy filter. Confirm on the vendor’s own privacy page before a regulated workload.
| Policy |
Typical meaning |
Examples in this catalog |
| Zero retention |
Provider states prompts are not kept after inference |
Groq, Together, Fireworks, DeepInfra, Vertex (as listed), Azure, Bedrock, Cerebras, Phala |
| 30-day retention |
Logs kept ~30 days |
Anthropic, Mistral, Meta, SpaceXAI/xAI, Cohere, Claude on AWS |
| Retains prompts |
Longer or unspecified retention |
OpenAI, DeepSeek, NVIDIA, Google AI Studio (55 days), MiniMax |
| Trains = Yes |
May train on API data (OpenRouter flag) |
NVIDIA, DeepSeek, Thinking Machines, Liquid |
| BYOK |
Bring-your-own-key on OpenRouter |
Most high-volume hosts; some small hosts are No |
EU/EEA HQ on the live table includes Mistral (France), Nebius (Netherlands), NextBit (Spain), Inceptron (Sweden). HQ ≠ data residency. For GDPR routing, use EUrouter and still filter providers.
Full table: providers-index.md
Reference files
By modality (start here for “which model?”):
- chat-agent-models.md — agent/chat models with about-text, context, tools, pricing, OpenRouter + vendor links
- image-models.md — image generation/editing
- video-models.md — video generation
- audio-music-models.md — music (Lyria), TTS, realtime audio, transcription
By provider / host:
- providers-index.md — every OpenRouter-listed provider (slug, HQ, trains, retention, BYOK, volume, model count)
- models-by-provider.md — complete OpenRouter model ID list per lab, cloud, and host
- labs.md — first-party model creators
- cloud-platforms.md — Azure, Bedrock, Vertex, Alibaba, Tencent, Baidu, Cloudflare, GPU clouds
- inference-hosts.md — Groq, Together, Fireworks, DeepInfra, Cerebras, and other hosts
- specialized.md — embeddings, apply/morph, confidential inference (image/video/speech details live in the modality files)
- routers.md — OpenRouter, EUrouter, LiteLLM, Portkey, Vercel, Cloudflare, Kong, Helicone, Requesty, TrueFoundry, AISIX
When the user asks which models a provider hosts or which model for chat/image/video/music, read the matching file above. Do not guess IDs from memory.
Entry template (used in grouped files)
Each researched provider records: OpenRouter slug, HQ/policy, official docs + pricing URLs, API base URL / env var, OpenRouter model IDs, a dated flagship price snapshot, and when to pick it.
1---2name: ai-providers3description: Knowledgebase of current AI API providers and models (chat/agent, image, video, music/speech), documentation links, pricing, privacy/retention, and routers/gateways. Use when selecting models or vendors, comparing OpenAI, Anthropic, Google, Mistral, DeepSeek, xAI/SpaceXAI, Groq, Together, Fireworks, Vertex, Bedrock, Azure, OpenRouter, EUrouter, LiteLLM, Portkey, Sora, Veo, FLUX, Lyria, Kling, Runway, or when the user asks which model to use for chat, agents, image, video, or music.4---56# AI Providers Knowledgebase78Catalog of **first-party labs**, **cloud platforms**, **inference hosts**, **specialized APIs**, and **routers/gateways**. Snapshot dated **September 2026**.910This skill is for **selection and comparison**. For SDK code, use the vendor skill:1112| Topic | Skill |13| :--- | :--- |14| OpenAI Responses / Chat Completions | [openai-sdk](../openai-sdk/SKILL.md) |15| Anthropic Messages / Claude | [anthropic-sdk](../anthropic-sdk/SKILL.md) |16| Gemini / Vertex via Gen AI SDK | [google-genai](../google-genai/SKILL.md) |17| Mistral / Codestral | [mistral-sdk](../mistral-sdk/SKILL.md) |18| Cohere Command / Embed / Rerank | [cohere-sdk](../cohere-sdk/SKILL.md) |19| DeepSeek OpenAI-compat | [deepseek-sdk](../deepseek-sdk/SKILL.md) |20| OpenRouter / EUrouter integration | [ai-routers](../ai-routers/SKILL.md) |2122## Freshness rule2324Prices, model IDs, and retention policies change without notice.25261. Treat numbers in this skill as a **dated snapshot** (Sep 2026).272. Before quoting a production price or pinning a model ID, re-fetch the **official pricing/docs URL** in the provider entry.283. Prefer the **vendor’s own docs** over OpenRouter for first-party APIs. Prefer OpenRouter’s live listing for *which hosts currently serve a model*.294. Do not invent model IDs from training memory. If this catalog and a live page disagree, use the live page.3031Live marketplace: [OpenRouter Providers](https://openrouter.ai/providers) · [OpenRouter Models](https://openrouter.ai/models)3233## When to use which path3435```36Need first-party features (Responses, Computer Use, native tools, enterprise DPA)37 → Call the lab or its cloud SKU (Azure / Vertex / Bedrock).3839Need lowest latency on open weights40 → Groq (LPU) or Cerebras (wafer-scale); SambaNova for high TPS.4142Need cheapest tokens on a popular open model43 → Compare DeepInfra, Novita, SiliconFlow, Parasail on OpenRouter for that model ID.4445Need one key, many labs, automatic failover46 → Hosted router: OpenRouter (global), EUrouter (EU residency), Vercel AI Gateway (Vercel apps).4748Need VPC, virtual keys, budgets, guardrails on *your* provider keys49 → Self-hosted / enterprise gateway: LiteLLM, Portkey, Kong, TrueFoundry.5051Need a chat / coding / tool-calling agent52 → chat-agent-models.md (GPT-6, Claude 5, Gemini 3.x, Grok, DeepSeek, Kimi, GLM, …).5354Need image generation or editing55 → image-models.md (GPT Image, FLUX.2, Nano Banana, Recraft, Seedream, Grok Imagine, Krea).5657Need video generation58 → video-models.md (Sora 2 Pro, Veo 3.1, Runway, Kling, Hailuo, Wan, Seedance, HeyGen).5960Need music or speech61 → audio-music-models.md (Lyria 3 music; Fish/MiniMax/Qwen/Deepgram TTS; gpt-audio; ASR).62```6364## Privacy cheat sheet (OpenRouter-reported, Sep 2026)6566OpenRouter’s provider table is the fastest policy filter. Confirm on the vendor’s own privacy page before a regulated workload.6768| Policy | Typical meaning | Examples in this catalog |69| :--- | :--- | :--- |70| **Zero retention** | Provider states prompts are not kept after inference | Groq, Together, Fireworks, DeepInfra, Vertex (as listed), Azure, Bedrock, Cerebras, Phala |71| **30-day retention** | Logs kept ~30 days | Anthropic, Mistral, Meta, SpaceXAI/xAI, Cohere, Claude on AWS |72| **Retains prompts** | Longer or unspecified retention | OpenAI, DeepSeek, NVIDIA, Google AI Studio (55 days), MiniMax |73| **Trains = Yes** | May train on API data (OpenRouter flag) | NVIDIA, DeepSeek, Thinking Machines, Liquid |74| **BYOK** | Bring-your-own-key on OpenRouter | Most high-volume hosts; some small hosts are No |7576EU/EEA HQ on the live table includes **Mistral (France)**, **Nebius (Netherlands)**, **NextBit (Spain)**, **Inceptron (Sweden)**. HQ ≠ data residency. For GDPR routing, use [EUrouter](https://www.eurouter.ai/docs) and still filter providers.7778Full table: [providers-index.md](references/providers-index.md)7980## Reference files8182**By modality (start here for “which model?”):**8384- [chat-agent-models.md](references/chat-agent-models.md) — agent/chat models with about-text, context, tools, pricing, OpenRouter + vendor links85- [image-models.md](references/image-models.md) — image generation/editing86- [video-models.md](references/video-models.md) — video generation87- [audio-music-models.md](references/audio-music-models.md) — music (Lyria), TTS, realtime audio, transcription8889**By provider / host:**9091- [providers-index.md](references/providers-index.md) — every OpenRouter-listed provider (slug, HQ, trains, retention, BYOK, volume, model count)92- [models-by-provider.md](references/models-by-provider.md) — complete OpenRouter model ID list per lab, cloud, and host93- [labs.md](references/labs.md) — first-party model creators94- [cloud-platforms.md](references/cloud-platforms.md) — Azure, Bedrock, Vertex, Alibaba, Tencent, Baidu, Cloudflare, GPU clouds95- [inference-hosts.md](references/inference-hosts.md) — Groq, Together, Fireworks, DeepInfra, Cerebras, and other hosts96- [specialized.md](references/specialized.md) — embeddings, apply/morph, confidential inference (image/video/speech details live in the modality files)97- [routers.md](references/routers.md) — OpenRouter, EUrouter, LiteLLM, Portkey, Vercel, Cloudflare, Kong, Helicone, Requesty, TrueFoundry, AISIX9899When the user asks **which models a provider hosts** or **which model for chat/image/video/music**, read the matching file above. Do not guess IDs from memory.100101## Entry template (used in grouped files)102103Each researched provider records: OpenRouter slug, HQ/policy, official docs + pricing URLs, API base URL / env var, **OpenRouter model IDs**, a **dated** flagship price snapshot, and when to pick it.
MatthewTaormina