Perf is a single API endpoint that automatically picks the best LLM for each request and guarantees valid JSON output.
You call perf.chat(messages) and it handles everything - classifies the task, routes to the cheapest model that can handle it (GPT-4o, Claude, Haiku, etc.), validates the output, and retries if needed.
No more JSON parsing failures. No more overpaying for simple tasks. No more writing routing logic yourself.
Screenshots
Product Updates (0)
No updates yet. Check back later for updates from the team.
Comments (1)
Ran 500 prompts through Perf: 100% valid JSON, 90% cheaper than GPT-4o baseline. Excited to get this out to those tired of JSON validation errors.