GitHub Models
#8 v LLM API a gatewaygithub · od 2024-08-01 (öffentliche Ankündigung/Beta); Public Preview ab 2024-10-29 · 18× · naposledy 30. 6. 2026
53
Momentum
GitHub Models is a cloud service operated by GitHub (Microsoft) that gives all GitHub users rate-limited free API access to over 45 AI models from multiple providers through a unified interface. An interactive playground allows prompt testing, comparison, and evaluation; models can then be integrated into applications via REST API or SDK. Beyond the free rate-limited tier, paid pay-as-you-go usage and BYOK (Bring Your Own Key) for OpenAI and Azure AI are available. The service runs entirely on GitHub/Azure infrastructure with no self-hosting option.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Deployment (Self-Hosted/Cloud) | Cloud only (GitHub/Azure infrastructure); no self-hosting; data is not used for training |
| Throughput/Latency | Free tier: ~10 RPM; high-tier models: 50 RPD, low-tier: 150 RPD; token limits vary by model (e.g., GPT-4.1: max. 1,048,576 input tokens, 32,768 output tokens in paid tier); Paid: Azure Foundry production limits |
| License | Proprietary service (GitHub/Microsoft); individual model licenses per publisher; privacy: data not used for training |
| Platform | Cloud (github.com/marketplace/models, API endpoint: models.github.ai); browser playground, integration in GitHub Codespaces & VS Code |
| Price | Free (rate-limited for all GitHub users); Paid: $0.00001 USD per token unit (pay-as-you-go); BYOK via own OpenAI/Azure keys (billed by provider) |
| Protocol Compatibility | REST API (models.github.ai/inference/chat/completions); Azure AI Inference SDK; OpenAI SDK (compatible); GitHub Personal Access Token (models:read) as auth |
| Release Date | August 1, 2024 (announcement/beta); Public Preview: October 29, 2024 (GitHub Universe) |
| Supported Models/Providers | 45+ models (2025); incl. OpenAI (GPT-4.1, GPT-4o, o3), Anthropic (Claude), Meta (Llama 3.x/4), Microsoft (Phi), DeepSeek, Cohere, AI21 (Jamba), Mistral; BYOK: OpenAI & Azure AI |