Where to get an API key for each provider

Every provider gives you a key from its own dashboard and bills you directly for what you use. The full list with direct links:

• OpenAIGPT-6 Astra, GPT-6.1 Sol and the rest of the GPT family.
• AnthropicClaude Opus 5.5, Sonnet 5.5, Fable 5.1 and Haiku 4.5.
• GoogleGemini 3.8 Flash, Gemini 3.1 Pro and Flash-Lite.
• xAIGrok 4.7, Grok 4.6 and the Grok 4.20 reasoning models.
• DeepSeekDeepSeek chat and reasoner, always pointing at the newest release.
• MistralMistral Large, Medium, Magistral reasoning and Codestral.
• GroqOpen models on LPU hardware. Often the fastest panel on screen.
• OpenRouterOne key for hundreds of models from every major lab.
• PerplexitySonar models that search the web before they answer.
• Together AIHosted open models: Llama, DeepSeek, Qwen and more.
• FireworksFast inference for open models. Sync to see the full catalog.
• CerebrasWafer-scale inference measured in thousands of tokens a second.
• MoonshotKimi K2 models with long context and strong coding.
• QwenAlibaba Qwen models through Model Studio (international).
• MetaLlama 4 Maverick, Llama 4 Scout and Llama 3.3 from the Meta Llama API.
• NVIDIANemotron 3 models on NVIDIA NIM, plus open models from other labs.
• CohereCommand A models built for business writing, search and tools.
• Hugging FaceOne token for open models from Meta, Google, NVIDIA, Microsoft and more.
• SambaNovaDeepSeek, Llama and gpt-oss on fast RDU chips. Free tier, no card.
• SiliconFlowOpen models with a set of always-free ones, like Qwen3 8B and GLM-4 9B.
• Z.aiGLM models from Zhipu. The Flash models are free, one request at a time.
• Ollama CloudBig open models like Kimi, GLM and DeepSeek on Ollama servers. Free plan for light use.
• Vercel AI GatewayOne key for GPT, Claude, Gemini and 400+ models, with $5 of free credit every 30 days.

After you add a key, press Sync all models in a panel's model menu to load every model that key can use, including models released after this page was written.