June 2026 monthly summary for braintrust-proxy. Delivered a targeted fix for Vertex-related 404s when invoking newer Gemini models by implementing complete model path mappings and aligning provider entries. Updated the SDK to construct correct Vertex requests using full publisher model paths, and expanded the model list to include missing full-path variants. Local validation confirms invoke() works for four Gemini models against Vertex and Playground without requiring test changes. Result: increased reliability, reduced incidence of model-invocation errors, and smoother developer experience when leveraging Gemini models via Vertex. Tech stack focus included mapping/configuration, SDK request construction, and Vertex gateway integration.