Simulated results | Pass rates with intervals | Compare the targets case by case | Where the targets fail | How many test cases a comparison needs | Check the judge before trusting it | Weigh quality against cost and latency | Report the method

Prepare requests | Submit and wait with the low-level API | Submit now and collect later | Clean up files

Endpoint and authentication families | OpenAI-compatible and project surfaces | Evaluations | Speech family | Content Safety surfaces | Current omissions | Compatibility policy

Configure a Content Safety resource | Moderate open-text answers | Add study-specific blocklists | Check whether an answer is grounded | Screen prompts and retrieved documents | Other Content Safety checks | Use the gate outputs

What embeddings are for | Generate embeddings | Compare same-source and cross-source pairs | Rank documents for a query | Cluster text responses | Work with larger collections

Why put embeddings in a recipe? | Create an embedding recipe | Fit a small classifier with PCA | Control the embedding step | Resampling, caching, and cost | Troubleshooting | Next steps

Setup | Grade a model on labeled examples | Audit labels already in your data | Compare two prompts in one evaluation | Evaluate an agent | Evaluate responses your application already stored | Safety evaluators | Finding the results later

Start with a versioned codebook | Extract only the coded fields | Compare with hand codes on a sample | Check repeated-run stability | Estimate a theme share with an interval | Record provenance | What to report

Configure image resources | Generate a research stimulus | Save and display the image | Make a controlled variant | Record what you used

Install | Configure credentials | Understand endpoints and routes | Know deployment names | Get a first response | Extract structured fields | Embed and compare text | Choose the next article

Response text and usage | Stateful turns | Structured extraction | User-defined R tools | Remote MCP tools | Web-grounded answers | Reasoning token accounting | Streaming and chat completions

Fast transcription | MAI-Transcribe enhanced mode | Transcribe non-English audio | Audio translation endpoints | Code the transcript | Synthesize speech | Notes for analysis