Jobs · 05
The model behind the feature
Use this when your app calls a model directly. Ranked for indie quality of life: docs, how you get a key, and how readable the bill is. Token rates change — check the provider site.
Initial editorial estimate. Not a measured benchmark. Edit scores in data/tools.json.
01
Best for Calling Claude from your own product
Claude models via API. The pair to Claude Code.
- + Claude is a default choice for coding and careful tool use
- + The API is straightforward to call from the AI SDK and most frameworks
- − You are on one lab's roadmap and rate limits
- − Rates: check site
02
Best for The model API most libraries already speak
The default model API. GPT models, tools, and a platform console.
- + The widest support in frameworks and examples
- + Usage-based, so a small app can start small
- − The bill tracks usage, and usage on agents adds up
- − Rates, model names, and limits: check site
03
Best for Switching models without switching SDKs
One API key for many models, including labs and open weights.
- + One account instead of five provider consoles
- + Model choice stays a config change
- − A middleman means another status page and another markup
- − Per-model rates: check site
04
Best for Low-latency calls to open models
Fast inference API for open models.
- + Speed is the reason to pick it
- + An OpenAI-compatible style API is part of the draw
- − The model menu is not the same as OpenAI or Anthropic flagships
- − Rates and free credits: check site
05
Best for A capable model API with a free starting quota
Google's Gemini models, with a free tier that moves.
- + A free tier has been part of the offer
- + Multimodal models are a first-class part of the API
- − Free-tier quotas are not a production plan
- − Model names and prices move. Check site
06
Best for A European model API, or open weights you can host
European model lab with an API and open-weight releases.
- + API and open weights are both real options
- + A credible alternative to the three biggest US labs
- − Ecosystem examples still assume OpenAI first
- − Rates and which models are on the API: check site